#23800: fix: map HTTP 404 to model_not_found failover reason

#### Summary HTTP 404 errors from model providers (e.g. Google Gemini's "Requested entity was not found") break the failover chain. `resolveFailoverReasonFromError()` handles 402/429/401/403/408/503/400 but **not 404**, so bare 404s propagate as unrecoverable errors instead of triggering model fallback. Closes #4992 Supersedes #11446 (irreconcilable branch conflicts after upstream added `model_not_found`) References #4997, #6735, #9427, #10262, #14510 (prior attempts, all closed unmerged) lobster-biscuit #### Root Cause Since upstream added `model_not_found` as a `FailoverReason` (with `resolveFailoverStatus` mapping and `isModelNotFoundErrorMessage` classification), most model-not-found scenarios are covered. However, `resolveFailoverReasonFromError()` still lacks a `status === 404` case. When a provider returns HTTP 404 with a message that doesn't contain "model" (e.g. Google's "Requested entity was not found"), message-based classification misses it and the failover chain breaks. #### Fix Single line: `if (status === 404) return "model_not_found"` in `resolveFailoverReasonFromError()`. Uses the existing `model_not_found` reason — no new types needed. #### False-Positive Risk Low. `resolveFailoverReasonFromError()` is only called from `coerceToFailoverError()` in the model fallback loop catch block. Only LLM provider HTTP errors flow through this path — file/config/module errors use different error codes (ENOENT, ERR_MODULE_NOT_FOUND). #### Tests - `failover-error.test.ts`: 3 new cases (404 → model_not_found, provider entity error, coercion with metadata) - `model-fallback.test.ts`: 1 integration case (bare 404 from primary → cascade to fallback) ``` pnpm vitest run src/agents/failover-error.test.ts src/agents/model-fallback.test.ts # Test Files 2 passed (2) # Tests 37 passed (37) ``` TypeScript: `pnpm tsgo` — no new errors Lint: `pnpm check` — clean #### Files Changed | File | Change | |------|--------| | `src/agents/failover-error.ts` | Add `status === 404 → "model_not_found"` in `resolveFailoverReasonFromError()` | | `src/agents/failover-error.test.ts` | 3 new test cases | | `src/agents/model-fallback.test.ts` | 1 integration test case | #### Sign-Off - Models used: Claude Opus 4.6 (implementation) + ChatGPT 5.3 Codex (verification) - Testing: Fully tested — 37/37 pass, no regressions - Verification: ChatGPT 5.3 Codex independently traced the full failover pipeline through `model-fallback.ts` → `coerceToFailoverError()` → `resolveFailoverReasonFromError()`, confirmed the gap, validated the fix approach, and verified false-positive risk is low since only LLM provider HTTP errors flow through this path - The fix is intentionally minimal: 1 line of logic + 4 test cases. Previous attempts (#4997, #6735, #9427, #10262) each missed at least one of the four gaps in the pipeline; upstream has since closed three of those gaps, leaving only this status-code mapping. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> Co-Verified-By: ChatGPT 5.3 Codex <noreply@openai.com>