feat(cognition,#1385): generate_response PR-2 — async evaluate_response + cognition/generate-response IPC handler

[joelteply]

Summary

Stacks on PR-1 #1388 (pure types + prompt builder + identity-reminder template, MERGED at 872e84a). PR-2 wires the async path: build_response_messages → adapter.generate_text (existing local Qwen router via global_registry) → result with timing + tokio::time::timeout replacing the TS Promise.race.

What this ships

• evaluate_response(GenerateResponseRequest) -> Result<GenerateResponseResult, GenerateResponseError> — async composer. Honors per-request model/temperature/max_tokens/timeout_ms overrides; defaults match TS (Qwen3.5 / 0.7 / 150 / 180_000ms).
• GenerateResponseError — typed: NoAdapter, Generation, Timeout. No silent default-on-error; caller picks fail-open vs fail-closed.
• build_response_generation_request(&request, model, start_ms) -> TextGenerationRequest — pure helper. Pins wire shape (provider="local", response_format=Text, purpose="cognition/generate-response", persona/room attribution).
• result_from_response(response, model, start_ms, end_ms) -> GenerateResponseResult — pure helper. Trims text, stamps model + timing, populates tokens_used only when total_tokens > 0 (mirrors TS truthiness check on usage object).
• cognition/generate-response command arm in CognitionModule::handle_command.

Discipline

• tokio::time::timeout wraps adapter.generate_text — clean Timeout variant on error enum (TS Promise.race equivalent).
• Saturating subtraction on response_time_ms — clock-backwards artifact (NTP adjustment mid-call) reports 0, not a wrapped huge u64.
• tokens_used = None when provider reports zeros — avoids emitting fake {0,0,0} measurements for providers that don't instrument usage.
• response_format=Text (TS default) — local Qwen takes plain text, no JSON-mode constraint.
• All constants documented with their TS-default origins.

Tests (10 new — full module now 39 passing)

build_response_generation_request (3): defaults shape (provider/model/temp/max/response_format/purpose/attribution/message count), overrides honored, caller timestamp embedded in identity reminder.

result_from_response (5): trims whitespace, stamps model + timing, populates tokens when provider reports total > 0, tokens None when zero, response_time saturates on clock-backwards.

GenerateResponseError (2): NoAdapter Display carries provider + model, Timeout Display includes duration.

Full cognition regression: 335/335 pass.

NOT in this PR

• PR-3: TS shim — AIDecisionService.generateResponse delegates to RustCoreIPCClient.cognitionGenerateResponse + cognition mixin binding.
• PR-4: Delete dead TS — buildResponseMessages helper + inline identity-reminder template (~250 LOC removed). After PR-3 + PR-4, AIDecisionService.ts is pure slot-coordination + IPC shim code.

Refs

• oxidizer: migrate AIDecisionService.generateResponse to Rust cognition/generate-response #1385 — sub-card
• feat(cognition,#1385): generate_response PR-1 — pure types + prompt builder + identity-reminder template #1388 — PR-1 (MERGED)
• TS-side AI logic that violates 'TS is thin glue' directive — umbrella triage #1248 — umbrella: TS-side AI logic violates 'TS is thin glue' directive
• Sibling pattern: oxidizer: migrate AIDecisionService.checkRedundancy to Rust cognition/check-redundancy #1375 check_redundancy stack (MERGED PRs feat(cognition,#1375): check_redundancy PR-1 — pure types + prompt + parser #1377/feat(cognition): wire check redundancy IPC #1381/feat(cognition): delegate redundancy shim to rust #1383)

Test plan

• cargo test --package continuum-core --lib --features metal,accelerate cognition::generate_response:: — 39/39 pass
• cargo test --package continuum-core --lib --features metal,accelerate cognition:: — 335/335 pass
• Pre-push TS clean, ESLint baseline held, Rust compile clean
• CI green
• PR-3 (TS shim) + PR-4 (dead-TS delete) stack on top

🤖 Generated with Claude Code