[2Zest]

Closes #71941

Summary

DataFrame.to_sql documents timedelta values as nanoseconds, but the SQL insertion path views the native timedelta buffer directly. Since pandas 3 uses timedelta64[us] by default, values were written 1,000 times too small.

The insertion path now converts NumPy-backed timedeltas to timedelta64[ns] before extracting integer values. Arrow-backed timedeltas already used the nanosecond conversion and remain unchanged.

Tests

• python -m compileall -q pandas/io/sql.py pandas/tests/io/test_sql.py
• python -m pytest pandas/tests/io/test_sql.py -k test_to_sql_timedelta_writes_nanoseconds -q (blocked locally because the in-place pandas extension module pandas._libs.pandas_parser is not built)
• git diff --check

I used OpenAI Codex to develop this pull request, reviewed and understood every change, and included the regression test.

• closes BUG: to_sql writes timedelta64 columns in their own unit instead of nanoseconds (1000x too small with the default "us") #71941
• Tests added and passed if fixing a bug or adding a new feature
• All code checks passed.
• Added type annotations to new arguments/methods/functions (no new arguments/methods/functions).
• Added an entry in the latest whatsnew file if fixing a bug or adding a new feature.
• I did not use AI to develop this pull request.
• I used AI to develop this pull request. I prompted it to follow the repository contribution guidance, reviewed and understood every change, and described above how I used it and exactly which tool and model.

[github-actions]

Thanks for the pull request, @2Zest! It's linked to #71941, but that issue isn't assigned to you yet. To make sure two people don't unknowingly work on the same thing, we ask contributors to claim an issue first. You can comment /take on #71941 to claim it, but be aware that certain issues cannot be taken if they are labeled e.g. Needs Triage — see the contributing guide for more details.

This pull request is unlikely to be reviewed until #71941 is assigned to you.