fix(agent): eager fallback on persistent transport failures (#22277)

[teknium1]

Summary

A degraded primary whose stream hangs (heartbeats keep the socket alive) now fails over to the configured fallback chain after one retry, instead of silently re-hitting the dead primary for 15+ minutes until the retry budget exhausts.

Root cause: the eager-fallback gate in conversation_loop.py fired only on rate_limit/billing. A stale-detector-killed hung stream classifies as FailoverReason.timeout (retryable=True) and fell through to retry-with-backoff against the same broken provider — 3 × ~180–300s stale kills + exponential backoff ≈ the reported 15+ min hang, with the fallback chain idle the whole time.

Changes

• agent/conversation_loop.py: extend the existing eager-fallback gate to also cover FailoverReason.timeout and FailoverReason.overloaded, but only after one real retry (retry_count >= 2) so genuine transient hiccups still recover on the primary. Reuses the same credential-pool-recovery guard and state reset as the rate-limit branch. No new config flag.
• scripts/release.py: AUTHOR_MAP entry for @linyubin.

Why this shape (not a new config flag, not first-stale failover)

The original eager-fallback feature (#1730) was deliberately scoped to rate limits — a 429 can't clear within the retry window, so retrying is pure waste. Timeout was left to retry-with-backoff on purpose, because a transport blip usually does recover on retry. This PR preserves that intent: it only fails over once a transport error has proven persistent (second attempt), so the common transient case keeps retrying the primary. Users with no fallback chain are unaffected.

Validation

Scenario	Behavior
timeout, attempt 1 fails	retry primary (transient-recovery intent preserved)
timeout, attempt 2 fails	activate fallback (caps bleed at ~2 stale cycles)
rate_limit	fires immediately (unchanged)
timeout, no fallback chain configured	keeps retrying primary (unchanged)

E2E verified with real imports (classify_api_error + gate logic). Targeted suites green: test_credential_pool_routing (10), test_error_classifier + test_failover_identity + test_gemini_fast_fallback (182). Total 192 tests.

Salvaged from PR #50228 by @linyubin, with their authorship preserved. Dropped #50228's error_classifier.py hunk (set should_fallback=True on the timeout result but the gate keys off the reason enum, never the flag — dead code).

Closes #22277.

Infographic

[github-actions]

🔎 Lint report: hermes/hermes-37d63a82 vs origin/main

ruff

Total: 0 on HEAD, 0 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 0 pre-existing issues carried over.

ty (type checker)

Total: 11522 on HEAD, 11520 on base (🆕 +2)

🆕 New issues (2):

Rule	Count
unresolved-attribute	2

First entries

run_agent.py:3014: [unresolved-attribute] unresolved-attribute: Object of type `Self@get_credits_spent_micros` has no attribute `_credits_session_start_micros`
tests/run_agent/test_credits_notices_toggle.py:76: [unresolved-attribute] unresolved-attribute: Unresolved attribute `_credits_session_start_micros` on type `AIAgent`

✅ Fixed issues (1):

Rule	Count
invalid-assignment	1

First entries

tests/run_agent/test_credits_notices_toggle.py:76: [invalid-assignment] invalid-assignment: Object of type `None` is not assignable to attribute `_credits_session_start_micros` of type `int`

Unchanged: 6065 pre-existing issues carried over.

Diagnostics are surfaced as warnings — this check never fails the build.