fix: keep year-search outages inside worker loop

[cursor]

Summary

• Prevent Radarr/Sonarr year discovery failures from escaping run_search_loop and terminating the search worker.
• Restore reuse of the year list across loop iterations, avoiding a full Arr library scan on every iteration.
• Add a regression test proving an Arr API outage triggers the standard 300-second in-process backoff instead of a critical worker crash.

Testing

• Added or updated automated tests
• Performed manual verification (describe below)
• Not applicable (explain why)

Manual test notes:

QBITRR_OVERRIDES_DATA_PATH=/workspace/.config python3 -m unittest discover -s tests — 394 passed. python3 -m pre_commit run --files qBitrr/arss/arr_base.py tests/test_search_and_db_query_hooks.py — passed.

Checklist

• Linked the related issue or discussed scope with maintainers
• Ran linters / formatting tools relevant to the touched code
• Updated documentation or release notes where it makes sense
• Confirmed there are no sensitive secrets or credentials committed

Additional Notes

No documentation change is needed: this restores the documented year-search behavior and existing outage-recovery contract. The regression was introduced by PR #524 when get_year_search() moved to every loop iteration outside the recovery boundary. With default SearchByYear=true, an Arr outage could exhaust the default five process restarts in 300 seconds and leave automated searches stopped until qBitrr restarted.