fix: use active checkout for update checks

[bxclawbot-source]

Summary

• prefer the running checkout for Hermes update/version checks
• invalidate stale cached behind counts when repo path or HEAD changes
• add regression coverage for repo selection and cache reuse

Testing

• python -m pytest tests/hermes_cli/test_update_check.py -q

[bxclawbot-source]

Updated branch to current fix commit 7e1b686 after replaying the same update-check patch onto the current main.

Local verification:

• python -m pytest tests/hermes_cli/test_update_check.py -q
• hermes --version now reports the active checkout correctly

Also switched local Hermes config to Anthropic + claude-sonnet-4.6 for day-to-day use; that config change is local-only and not part of this PR.

[bxclawbot-source]

Added follow-up commit 09aa3c6c on top of the existing branch.

This extends the PR to fix a second issue: when fallback activated, the request runtime snapshot still pointed at the exhausted primary provider, so retries kept hitting the old backend.

Verification:

• python -m pytest tests/run_agent/test_provider_fallback.py -q
• forced local simulation confirms fallback now updates runtime provider/model/client kwargs before the next retry

[bxclawbot-source]

Added follow-up commit 0d33314 for the cheaper failover stack.

This keeps the request runtime snapshot aligned when fallback activates to OpenRouter / google/gemini-2.5-flash, so retries actually use the backup model instead of the exhausted primary.

Verification:

• python -m pytest tests/run_agent/test_provider_fallback.py -q
• forced local fallback simulation: anthropic -> openrouter/google/gemini-2.5-flash

[teknium1]

Closing as superseded by #22475.

Triage notes (medium confidence):
main hermes_cli/banner.py:253-256 and _resolve_repo_dir at line 270 already prefer Path(file).parent.parent over $HERMES_HOME/hermes-agent; merged PR #22475 'fix stale update-check repo resolution' addressed this.

Thanks for the contribution — the underlying problem this PR addresses has been resolved by the linked PR on current main. If you believe this was closed in error, please comment and we'll reopen.

(Bulk-closed during a CLI PR triage sweep.)