cl: preserve expanded Gloas lifecycle hardening

[domiwei]

Context

This draft preserves the broader Gloas hardening work extracted from #22683 so the original checkpoint-sync fix can remain small and mergeable.

It is intentionally stacked on #22683. Review should wait until that PR lands and this branch is rebased onto main.

Included follow-up clusters

• shared Engine payload-validation coordination and fork-choice lock lifecycle
• backward/forward Gloas lookahead and skipped-envelope recovery
• DA/EL retry ownership and live parent-envelope resolution
• durable envelope publication, index reconciliation, prune, and restart recovery
• related PeerDAS boundary hardening beyond the startup-order fix retained in cl: fix Gloas checkpoint sync with external execution clients #22683
• canonical SSZ rejection across network and disk ingress

Status

Draft only. The preserved branch contains the complete expanded implementation and tests, but it still needs decomposition and focused lifecycle work before review:

• resolver admission can queue longer than the scheduled child lifetime under saturation
• skipped-envelope recovery inherits a backlog-scaled retry budget that can hold completion for days at the configured cap

Those risks are intentionally quarantined here and are not part of #22683.

The next step is to split this holding PR into coherent dependency-ordered follow-ups rather than merge it as one unit.

[domiwei]

Carried review feedback from #22683

This draft preserves code removed from the slim PR; it does not imply that the removed implementation is ready to merge. Before any cluster is promoted into a follow-up PR, retain and re-evaluate the applicable review findings from #22683, including:

• trust-before-use for forward/backward lookahead and EMPTY/FULL inference
• peer-fault versus local-state-unavailable error classification
• nil-engine stage handoff and single-block forward progress
• repeated full-state copies and repeated block/envelope hashing
• bounded/fair HTTP and P2P recovery, capacity backpressure, and restart ownership
• envelope-before-block behavior and live resolver retry/expiry ownership
• prune progress, scan/fsync complexity, and startup scratch-artifact recovery
• observer aggregate CPU/allocation impact and PeerDAS lifecycle concerns beyond the startup-order fix retained in cl: fix Gloas checkpoint sync with external execution clients #22683

The inline threads on #22683 were resolved there because their referenced code is absent from the slim diff, not because these concerns were waived. This draft remains a holding branch and should be decomposed into dependency-ordered PRs with each applicable finding addressed and retested.