fix(storage): preserve WAL on checkpoint failure

[WangXukang-cypher]

Summary

• Preserve the DuckDB WAL file when checkpoint recovery fails instead of deleting it.
• Propagate checkpoint errors to prevent startup from continuing after an unsuccessful recovery.
• Ensure the temporary DuckDB connection is closed on both success and failure.

Deleting the WAL after a failed checkpoint can permanently discard committed transactions that have not yet been merged into the main database file.

Test plan

• Verified src/sirchmunk/storage/duckdb.py compiles with python3 -m py_compile.
• Verified checkpoint errors are propagated instead of being swallowed.
• Verified the WAL file and its contents are preserved when checkpoint recovery fails.
• Verified the temporary DuckDB connection is closed after a checkpoint failure.
• Passed trailing-whitespace and end-of-file checks.
• Passed the added-large-file check.
• Passed Ruff checks.
• Confirmed the repository currently has no pytest tests to collect.