BUG: fix to_csv chunksize producing inconsistent dt64/td64 formatting

[jbrockmendel]

Summary

• Fixes inconsistent datetime and timedelta formatting in DataFrame.to_csv when chunksize causes columns to be formatted chunk-by-chunk instead of using the full column data
• Pre-computes _is_dates_only for each datetime/timedelta column using the full column before chunking, and pre-formats the index when it is a DatetimeIndex/TimedeltaIndex
• Adds 6 tests covering datetime columns, timedelta columns, datetime index, multiple datetime columns, dates-only preservation, and chunked-vs-default equivalence

closes #55481

Test plan

• Existing CSV tests pass (672 total)
• mypy and pyright clean on modified file
• pre-commit passes
• New tests cover: datetime column, timedelta column, datetime index, multiple datetime columns with mixed dates-only, dates-only preservation, and chunked vs non-chunked equivalence

🤖 Generated with Claude Code