fix: emit status notification on credential pool rotation

[nightq]

Summary

Fixes inconsistent notification behavior when credential pool rotation causes a provider switch.

Root Cause

When fallback_model triggers a provider switch, Hermes correctly notifies the user via _emit_status(). However, when credential pool exhaustion causes the same provider switch, no notification was sent. The user had no idea they were talking to a different AI.

Fix

Add _emit_status() calls in _recover_with_credential_pool() when credentials are rotated:

• Billing exhaustion (402): "🔄 Credential exhausted (402) — rotated to: {entry_id}"
• Rate limit (429): "🔄 Rate limited (429) — rotated to: {entry_id}"
• Auth failure (401): "🔄 Auth failed (401) — rotated to: {entry_id}"

Also fixed _emit_status() to use getattr() for status_callback to handle edge cases where the attribute doesn't exist (e.g., in test agents).

Test Plan

• All credential pool routing tests pass
• Manual verification of status emission

Closes #10476

[teknium1]

Thanks for the focused report and implementation. This is an automated hermes-sweeper review; the linked fallback-notification behavior is now implemented on current main.

• agent/conversation_loop.py:3166-3207 escalates rate-limit/billing failures to the configured fallback chain when credential-pool recovery cannot proceed.
• agent/chat_completion_helpers.py:1544-1557 records a durable old-provider/model → new-provider/model notification whenever fallback activates.
• agent/conversation_loop.py:5065-5070 emits that notification after successful recovery, before clearing transient retry status; run_agent.py:984-1002 delivers it exactly once through _emit_status().
• tests/run_agent/test_retry_status_buffer.py:138-163 covers the successful fallback-notice path.
• This landed in 3fe7f6d27a005a0a75644c1d1d42cef8c8750c5f (fix: preserve fallback switch notice on successful fallback). The reviewed PR's original run_agent.py recovery body is now a forwarder to agent/agent_runtime_helpers.py:696-954.

Current main intentionally keeps successful same-provider credential rotations as logged retry recovery rather than immediate lifecycle chatter; the durable cross-provider/model switch requested by #10476 is surfaced.