perf: Limit the cache size for to_datetime

[reswqa]

Fix #15736.

s = pl.datetime_range(date(2000, 1, 1), date(2001, 1, 1), '1s', eager=True)
strings = s.dt.strftime('%Y-%m-%dT%H:%M:%S')

Without cache:

%timeit strings.str.to_datetime(format='%Y-%m-%dT%H:%M:%S', cache=False)
1.47 s ± 71.9 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

With cache:

%timeit strings.str.to_datetime(format='%Y-%m-%dT%H:%M:%S', cache=True)

• Before this PR

18.2 s ± 790 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

• After this PR

1.88 s ± 59.4 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

Consider that the test data are all unique(The cache will never hit) and large, I think the overhead is acceptable.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 80.38%. Comparing base (0c2783a) to head (dcd16f1).
Report is 30 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #15826      +/-   ##
==========================================
- Coverage   81.35%   80.38%   -0.97%     
==========================================
  Files        1379     1263     -116     
  Lines      176619   165368   -11251     
  Branches     2544        0    -2544     
==========================================
- Hits       143686   132933   -10753     
+ Misses      32449    32435      -14     
+ Partials      484        0     -484     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[codspeed-hq]

CodSpeed Performance Report

Merging #15826 will not alter performance

_{Comparing reswqa:to_date_cache (dcd16f1) with main (a078d0c)}

Summary

✅ 22 untouched benchmarks

[ritchie46]

Good find. Thanks @reswqa