Fix batched thumbnail cache clearing

[ewlarson]

Summary

• Process every resource ID supplied to clear_thumbnail_cache.py inside one async event loop.
• Preserve the existing single-ID/multi-ID CLI and cleared/total summary.
• Add regression coverage proving every ID uses the same running loop and the CLI invokes one batch coroutine.

Production incident

A Kamal production run attempted to clear thumbnail caches for 13 UNR resources:

python /app/backend/scripts/clear_thumbnail_cache.py RESOURCE_ID_1 RESOURCE_ID_2 ...

The first resource completed, but the second aborted with:

RuntimeError: Event loop is closed
Task ... got Future ... attached to a different loop

Root cause

main() called asyncio.run(clear_thumbnail_for_resource(rid)) once for every resource. Each call created and closed a new event loop, while the process-global SQLAlchemy async engine/pool retained asyncpg connections bound to the previous loop. Checking out one of those connections for the next resource triggered the cross-loop failure.

Fix

The resource loop now lives in clear_thumbnails_for_resources(), and main() calls asyncio.run() once for the whole batch. Exceptions retain the resource-specific log message and continue to propagate as before.

User impact

Multi-resource cache purge commands now process the full resource list rather than failing after the first ID. Single-resource behavior and CLI syntax are unchanged.

Validation

• BTAA_SKIP_TEST_DB=1 python -m pytest backend/tests/scripts/test_clear_thumbnail_cache.py -q — 2 passed
• ruff check passed
• ruff format --check passed
• git diff --check passed

Issue tracking

A standalone issue could not be created because GitHub Issues are disabled for ewlarson/ogm-api (GitHub API returned HTTP 410). This PR includes the complete issue write-up so the incident remains documented.