Fix atomic catalog writes for invalid SQL identifiers

[CAOShurong]

Closes #1459

Summary

• keep the catalog node insertion in the same transaction as data-source creation so identifier or storage initialization failures roll the node back
• return 422 for reserved, malformed, and forbidden SQL identifiers, matching the existing overlength-identifier behavior
• cover the failure path end to end, including proving that the key remains available for a corrected retry

Testing

• uv run pytest -n 0 tests/test_writing.py tests/adapters/test_sql.py tests/adapters/test_sql_arrays.py -q (183 passed, 61 skipped, 11 xfailed)
• uv run pytest -n 0 tests/test_catalog.py -q (40 passed, 29 skipped)
• uv run pre-commit run --files CHANGELOG.md tiled/adapters/sql.py tiled/catalog/adapter.py tests/test_writing.py

Checklist

• Added a changelog entry
• Added the issue number closed by this PR

AI assistance

OpenAI Codex was used to develop this change. The resulting diff and tests were reviewed and verified locally before submission.