task: Implement Kiro spec large-payload-streaming-fast-path

[matdev83]

Implement the approved Kiro SDD for #503 introduced by merged spec-only PR #531.

Spec: .kiro/specs/large-payload-streaming-fast-path/

Outcome

Deliver the regression-safe large-request fast path defined by the spec: bounded-memory pre-commit replayable capture/spooling, streaming JSON validation/metadata proof, explicit frontend/backend wire-compatibility contracts, exact-once secure-session/canonical fallback orchestration, token-aware bounded model rewrite, replay-safe retry/failover, and measured protocol certification.

The existing canonical request path is the behavioral oracle and fallback. Preserving existing proxy functionality is more important than retaining the optimization. Any uncertain request shape, feature authority, route, backend, accounting/guardrail requirement, or semantic mismatch must fall back to canonical execution.

The authoritative implementation order is tasks.md: 16 ordered workstreams / 52 explicit subtasks.

Start gate / current ordering

• spec: define regression-safe large-payload streaming fast path (#503) #531 is merged. The spec-introduction gate is resolved.
• task: Implement Kiro spec extension-plane-declaration-consolidation #451 is complete through docs(spec): archive extension plane consolidation #547 and task: Implement Kiro spec extension-plane-review-corrections #549 review corrections are complete through fix(extension-planes): apply review corrections #555/docs(spec): archive extension plane review corrections #556; those former start gates are historical baseline.
• There is currently no active hard prerequisite for this task inside OSS Base. Production implementation may begin against current main.
• Before changing production code, execute Task 1 exactly as written against current main: re-inventory landed request-body/extension-plane dependencies and freeze fresh characterization/performance baselines. This is required revalidation, not another issue dependency.

Current dependency order:

#532 this implementation task
        ↓
#503 feature tracker
        ↓
#398 OSS Base split gate

Keep this transitive. Do not add #532 directly as a blocker of #398 while #503 remains the feature-level gate.

Relationship to #394 performance hardening

• Do not block this task on audit(performance): Reality check for high volume, highly concurrent proxy use #394, and do not block audit(performance): Reality check for high volume, highly concurrent proxy use #394 on this task. audit(performance): Reality check for high volume, highly concurrent proxy use #394 remains the broad performance-audit/benchmark umbrella and can continue evidence/harness work independently.
• feat(performance): Add large-payload streaming fast path to avoid full request materialization #503 owns this concrete optimization and should use audit(performance): Reality check for high volume, highly concurrent proxy use #394's measurement discipline where applicable.
• If audit(performance): Reality check for high volume, highly concurrent proxy use #394 captures relevant large-request/allocation/GC/ingress baselines before this task lands, refresh those affected measurements after the certified fast path lands.

Relationship to newer simplification work

• Merged spec: define pre-OSS core slimming and feature ownership closure #557 is being implemented by OSS Base work order task: Implement Kiro spec pre-oss-core-slimming #571. Its production ownership waves through Task 8.3 have now landed in feat(pre-oss): closed-plane contract on generated planes (1.x+2.x) #576/feat(pre-oss): move tool-call repair into feature (3.x) #582/feat(pre-oss): move secret-guard into feature + compose adapter (4.x) #578/feat(pre-oss): relocate compaction detector to infra (5.x) #579/feat(pre-oss): runtimebundle adapters + zero feature imports (6.x) #580/feat(pre-oss): ownership ratchets, SDK fixture, docs, evidence (7.x+8.x) #581, with merged-main inventory verification repaired in chore(pre-oss): pin inventory SHAs to merged main #584/fix(pre-oss): baseline SHA validated as mainline ancestor #585/fix(pre-oss): hermetic git identity in inventory SHA test #586. task: Implement Kiro spec pre-oss-core-slimming #571 remains open only for final Task 8.4 certification. task: Implement Kiro spec pre-oss-core-slimming #571 is not a hard blocker of task: Implement Kiro spec large-payload-streaming-fast-path #532. Task 1 must use the current post-slimming package/ownership layout and must not recreate retired pre-feat(pre-oss): closed-plane contract on generated planes (1.x+2.x) #576 owners.
• Merged spec: define full core feature ownership closure #559 / post-split task task: Implement Kiro spec core-feature-ownership-full-closure #572 is follow-up full ownership closure. It is outside this milestone and must not become a cross-milestone dependency for task: Implement Kiro spec large-payload-streaming-fast-path #532.
• feat(cli): Add thin client CLI for proxy control/admin APIs #348 and feat(interactive-commands): Add support for in-session interactive commands #389 are independent product/control-plane work; no hard dependency in either direction.
• feat(rebranding): Complete project rebranding #429 rebranding is coordination only. Prefer avoiding overlap with this high-churn request/runtime work where practical, but do not create a dependency edge. If rebranding lands first, Task 1 must re-inventory renamed paths.

Non-negotiable delivery constraints

• Feature remains disabled by default in the first release.
• No upstream request-body byte is committed before complete validation and final eligibility/authority proof.
• Never fabricate a partial/minimal lipapi.Call; canonical code receives only a fully decoded/validated call.
• Never move route/failover ownership into frontends/backends or add provider-name switches in generic core.
• Do not prune/reorder failover/race candidates just to retain the optimization.
• Static/dynamic uncertainty is canonical fallback, not a new rejection.
• Secure-session/A-leg preparation remains exactly-once across post-identity canonical continuation.
• Preserve current JSON limits/trust-boundary, guardrail, billing/accounting, conversation-projection and extension-plane authority.
• Protocol certification remains sequential according to spec: define regression-safe large-payload streaming fast path (#503) #531.
• Keep implementation PRs focused/chronological rather than combining all subtasks into one giant change.

Completion evidence

Do not close this work order until all applicable tasks.md work is complete and evidenced, including canonical-path characterization, streaming JSON differential/fuzz/limit parity, replay/spool race and cleanup tests, exact-once session/A-leg fallback, route-wide wire compatibility, canonical-vs-wire certification, retry/failover parity, bounded observability/architecture ratchets, material multi-MiB allocation/GC improvement, forced-fallback equivalence, and full repository quality/parity/race/QA gates.

If a proposed certified lane does not demonstrate worthwhile ROI or cannot prove semantic parity, leave that lane canonical-only rather than weakening correctness.

Tracker relationship

• Implements feat(performance): Add large-payload streaming fast path to avoid full request materialization #503.
• Authoritative merged spec: spec: define regression-safe large-payload streaming fast path (#503) #531.
• No active hard blocker.
• feat(performance): Add large-payload streaming fast path to avoid full request materialization #503 remains the feature-level OSS Base tracker and direct prerequisite of feat(split): OSS / Enterprise split [ENDS OSS Base Milestone] ============================================ #398.
• Closing task: Implement Kiro spec large-payload-streaming-fast-path #532 permits feat(performance): Add large-payload streaming fast path to avoid full request materialization #503 to close only after feature-level acceptance/completion evidence has been checked.

[matdev83]

Status sync after #535–#538

The hard-start gates in this work order remain valid, but the upstream state has advanced materially:

• task: Implement Kiro spec extension-plane-declaration-consolidation #451 has now landed W0–W5 plus tasks 9.1–9.2 through merged feat(extension-planes): complete frozen and diagnostics projections #538. Its remaining canonical work is 9.3, 9.4, and 10.1–10.3, so it is still a genuine production-start blocker for this task.
• spec: define regression-safe large-payload streaming fast path (#503) #531 remains open/unmerged, although its canonical spec metadata is fully approved and ready_for_implementation: true.
• Therefore task: Implement Kiro spec large-payload-streaming-fast-path #532 remains correctly planned, not status:in-progress. Production implementation should start only after both gates resolve, followed by the required fresh Task-1 re-inventory on then-current main.

Dependency shape remains #451 → #532 → #503 → #398. #394 stays independent; do not add #451/#532 directly to #398.

[matdev83]

Status sync — pre-OSS ownership moves have landed; no dependency change

The body section saying #571's production core slimming has not yet landed is now stale. The implementation waves through Task 8.3 are on main via #576/#582/#578/#579/#580/#581, with merged-main inventory fixes in #584/#585/#586. #571 remains open only for final Task 8.4 certification, including authoritative Linux race evidence.

This does not create a hard dependency for #532. #532 still has no active same-milestone prerequisite and may proceed independently. Its Task 1 re-inventory should now treat the post-slimming package/ownership layout on current main as the brownfield baseline and must not recreate pre-#576 owners. The intended dependency chain remains #532 → #503 → #398; do not add #532 directly to #398.