Summary

Fix eager fallback when the primary provider returns HTTP 402 / FailoverReason.billing.

The real bug was not in cron wiring. cron/scheduler.py already passes both
fallback_model and credential_pool into AIAgent. The failure was in the
eager-fallback guard: billing exhaustion was grouped with rate limits, then
_pool_may_recover_from_rate_limit() could still block fallback while waiting
for credential-pool rotation.

That is wrong for billing/account exhaustion. Rotating credentials should not
delay fallback for FailoverReason.billing.

Fixes #23138.

Changes

• teach _pool_may_recover_from_rate_limit() about the failover reason
• force it to return False for FailoverReason.billing
• pass classified.reason into the eager-fallback guard in run_conversation()
• add regression tests for:
  • helper-level billing/pool behavior
  • eager fallback with a multi-entry credential pool
  • run-level HTTP 402 -> immediate fallback

Why this fixes the issue

Before this change, a primary 402 Insufficient Balance could be treated as
"maybe credential rotation can recover", which prevented eager fallback from
activating even though fallback_providers was configured.

After this change, 402 -> FailoverReason.billing skips pool-rotation gating
and switches to the fallback provider immediately.

Validation

Passed:

• venv\Scripts\python.exe -m pytest tests/run_agent/test_run_agent.py -k billing_error_triggers_eager_fallback_before_retry_budget -q
• venv\Scripts\python.exe -m pytest tests/run_agent/test_provider_fallback.py tests/agent/test_credential_pool_routing.py tests/agent/test_error_classifier.py -q