feat: add non_existent arg to replace_time_zone

[MarcoGorelli]

closes #11579

needs rebasing onto #14961

This allows users to do:

In [1]: from datetime import datetime
   ...: 
   ...: import polars as pl
   ...: 
   ...: df = pl.DataFrame(
   ...:     {
   ...:         "ts": pl.datetime_range(
   ...:             datetime(2020, 3, 29, 0), datetime(2020, 3, 29, 5), "1h", eager=True
   ...:         ),
   ...:     },
   ...: )
   ...: 
   ...: df.with_columns(
   ...:     ts_london=pl.col("ts").dt.replace_time_zone("Europe/London", non_existent="null")
   ...: )
Out[1]: 
shape: (6, 2)
┌─────────────────────┬─────────────────────────────┐
│ ts                  ┆ ts_london                   │
│ ---                 ┆ ---                         │
│ datetime[μs]        ┆ datetime[μs, Europe/London] │
╞═════════════════════╪═════════════════════════════╡
│ 2020-03-29 00:00:00 ┆ 2020-03-29 00:00:00 GMT     │
│ 2020-03-29 01:00:00 ┆ null                        │
│ 2020-03-29 02:00:00 ┆ 2020-03-29 02:00:00 BST     │
│ 2020-03-29 03:00:00 ┆ 2020-03-29 03:00:00 BST     │
│ 2020-03-29 04:00:00 ┆ 2020-03-29 04:00:00 BST     │
│ 2020-03-29 05:00:00 ┆ 2020-03-29 05:00:00 BST     │
└─────────────────────┴─────────────────────────────┘

Currently there's no way to get that output, as non-existent datetimes just raise (you could convert to string, then use to_datetime with strict=False, but even then that would make all inputs which don't match the given format become null with no way to distinguish the non-existent ones to the invalid ones from those which didn't match the given format)

For now, I'm only adding it to replace_time_zone, and preserving current behaviour in other places. Adding non_existent to to_datetime would require quite a refactor because of how this would interact with strict, but I think for now it's valuable enough to add this to replace_time_zone

[codecov]

Codecov Report

Attention: Patch coverage is 98.51852% with 2 lines in your changes are missing coverage. Please review.

Project coverage is 81.07%. Comparing base (5d449cc) to head (8db1fd2).

Files	Patch %	Lines
crates/polars-arrow/src/legacy/kernels/time.rs	85.71%	1 Missing ⚠️
...ates/polars-plan/src/dsl/function_expr/datetime.rs	80.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #15062      +/-   ##
==========================================
+ Coverage   81.06%   81.07%   +0.01%     
==========================================
  Files        1342     1342              
  Lines      173935   174030      +95     
  Branches     2459     2459              
==========================================
+ Hits       141000   141097      +97     
+ Misses      32468    32467       -1     
+ Partials      467      466       -1     

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

[MarcoGorelli]

unrelated to this PR, but codecov flagged it as uncovered, so it's a good chance to add it

[deanm0000]

I with I stumbled on this before I did this monstrousity

df.join(
    df.select(
        ts=pl.datetime_range(
            pl.col('ts').first().dt.replace_time_zone("Europe/London").dt.convert_time_zone('UTC'), 
            pl.col('ts').last().dt.replace_time_zone("Europe/London").dt.convert_time_zone('UTC'), 
            '1h'
            )
        )
    .with_columns(
        ts_london=(lon:=pl.col('ts').dt.convert_time_zone('Europe/London')), 
        ts=lon.dt.replace_time_zone(None)
        ),
    on='ts', how='left')