PERF: Speed up import pandas with PEP 810 lazy imports

[eendebakpt]

PEP 810 (explicit lazy imports, CPython 3.15+) lets a module defer imports via a __lazy_modules__ declaration. On older Pythons the attribute is simply ignored, so adding it is backward compatible. A small set of pandas subsystems are imported eagerly today but are only needed once a specific entry point is called. Deferring them with __lazy_modules__ keeps them out of a plain import pandas:

• the I/O backends — read_* / to_* / Excel / parquet / html / sql / json /
  stata / sas / … and the optional third-party engines they pull in;
• pandas.testing (the assert_* helpers);
• the eval / query, resample, interchange, and pandas.api re-export stacks.

Make more imports lazy would require reworking pandas.core as the many dependencies in modules there make it hard to use PEP 810 without additional changes.

Benchmarks

The import time depends a bit on the workload, so adding multiple benchmarks.

workload (after import pandas)	baseline: modules / ms	lazy: modules / ms	faster
import pandas (nothing else)	290 / 826	221 / 678	18%
pd.DataFrame({'a':[1,2,3],'b':[4,5,6]})	290 / 838	221 / 682	19%
df.groupby('a').sum()	290 / 834	221 / 685	18%
pd.to_datetime([...])	290 / 836	221 / 704	16%
df.eval('a + b')	290 / 843	229 / 715	15%
df.to_csv()	291 / 832	222 / 706	15%
pd.read_csv(StringIO)	290 / 834	265 / 812	3%
pd.testing.assert_frame_equal(df, df)	290 / 840	229 / 710	15%

• [N/A] closes #xxxx (Replace xxxx with the GitHub issue number)
• Tests added and passed if fixing a bug or adding a new feature
• All code checks passed.
• [N/A] Added type annotations to new arguments/methods/functions.
• [N/A] Added an entry in the latest doc/source/whatsnew/vX.X.X.rst file if fixing a bug or adding a new feature.
• I have reviewed and followed all the contribution guidelines
• If I used AI to develop this pull request, I prompted it to follow AGENTS.md.