fix(worker): normalize encoded document filenames - #386

Summary

• decode percent-encoded URL basenames before storing URL-ingestion source filenames
• defensively decode legacy encoded filenames at the worker filesystem boundary
• add regression coverage for the production Chinese .doc filename pattern

Production evidence

This fixes the failure in Logfire trace 01a0770227cb4797dc15fe52b0b3ab15 (job_53f9d9d7b6d7). LibreOffice returned exit code 0 but wrote the .docx under the decoded Chinese output directory, while the worker checked for an encoded expected path. The same document succeeded in production when its decoded filename was used.

Verification

• uv run --project apps/worker pytest apps/worker/tests/unit/test_internal_parse_name.py -q — 2 passed
• uv run --project apps/api pytest apps/api/tests/unit/test_document_ingestion_filename.py -q — 1 passed
• focused Ruff checks — passed
• git diff --check — passed

The full worker unit suite was not runnable locally because required environment settings are absent (S3_BUCKET_NAME, S3_TEMP_PATH, DATABASE_URL, TMP_PATH).