feat(inference): inference-llm PR-4 — adapter integration (translation + new constructors)

[joelteply]

Summary

Bridges the substrate's typed InferenceRequest/InferenceComplete surface to the existing AIProviderAdapter trait (LlamaCppAdapter for local llama.cpp). Pure Rust, zero TS, zero shim. PR-5 ships the LlamaCppAdapter Runtime wiring + end-to-end stub-adapter test; PR-4 ships the translation logic + new constructors so PR-5 is just plumbing.

What lands

• InferenceRequest.prompt_text: Option<String> — wire addition for adapter-based engines that tokenize internally
• InferenceComplete.completion_text: Option<String> — wire addition for adapter-based engines that return text not tokens
• InferenceLlmModule.adapter: Option<Arc<dyn AIProviderAdapter>> + with_adapter / with_bus_and_adapter constructors
• handle_request routing: adapter when wired + prompt_text present; loud refusal when adapter wired + no prompt_text (raw-token path not yet implemented — never silent fallback); stub when no adapter
• run_adapter_inference — translates InferenceRequest → TextGenerationRequest, calls adapter, translates TextGenerationResponse → (InferenceComplete, FirstTokenEmitted)
• translate_adapter_response + translate_adapter_finish_reason — pure-function translation
• Cross-enum mapping: Stop→Stop, Length→MaxTokens, ToolUse→Error{reason} (loud refusal), Error→Error{reason}

Wire-shape decisions

• max_tokens=0 in substrate's GenerationBudget → None on adapter (substrate convention: 0=unlimited; adapter convention: None=unlimited, 0=stop-immediately which no caller wants)
• Empty stop_sequences → None on adapter
• persona_id propagates as stringified UUID for per-persona resource attribution
• purpose hardcoded "inference-llm" for adapter routing diagnostics

Sub-fix: missing TS bindings from PR-1

PR-1 (#1387) shipped the Rust types but the shared/generated/inference_llm/ directory of TS exports wasn't included in the commit (regen produced them locally; didn't stage). PR-4 ships all 10 TS files + the barrel index. Closes a wire-contract gap.

Test plan

• cargo test --lib --features metal,accelerate inference::llm_module — 44/44 pass (12 PR-3b + new PR-4):
  • translate_adapter_response_carries_text_and_usage — text + tokens_generated mapping
  • translate_finish_reason_covers_all_adapter_variants — cross-enum mapping pin
  • with_adapter_constructor_routes_via_adapter_path — constructors compile + no-adapter regression
• No regressions across other 2928 lib tests
• Pre-push gate clean

End-to-end "stub adapter via Arc<dyn AIProviderAdapter>" tests deferred to PR-5: the AIProviderAdapter trait has 8+ methods (provider_id / api_style / default_model / get_available_models / health_check / model_metadata / capabilities / etc) and implementing all on a test stub here would pull in ProviderHealth + AdapterCapabilities + ApiStyle + ModelInfo + their dependencies — bigger than atomic-slice. PR-5 will wire LlamaCppAdapter directly through Runtime registration.

Stack

• feat(inference): inference-llm PR-1 — typed event surface (MODULE-CATALOG §II) #1387 — inference-llm PR-1: typed event surface
• feat(inference): inference-llm PR-2 — InferenceLlmModule ServiceModule impl (stub-backed) #1391 — inference-llm PR-2: ServiceModule impl (stub-backed)
• feat(inference): inference-llm PR-3a — canonical ArtifactKeys + publishing helpers #1392 — inference-llm PR-3a: bus keys + publishing helpers
• feat(inference): inference-llm PR-3b — InferenceLlmModule auto-publishes via bus hook (pure Rust) #1393 — inference-llm PR-3b: auto-publish wiring
• This PR — inference-llm PR-4: adapter integration (translation + constructors)
• NEXT — PR-5: LlamaCppAdapter Runtime wiring + end-to-end test through real adapter

🤖 Generated with Claude Code