feat: let "ambiguous" take "null" value

[MarcoGorelli]

to be rebased onto #14958

Quick demo of what this lets you do:

In [13]: s = pl.datetime_range(datetime(2018, 10, 28), datetime(2018, 10, 28, 5), '1h', eager=True)
    ...: 

In [14]: s
Out[14]: 
shape: (6,)
Series: 'literal' [datetime[μs]]
[
        2018-10-28 00:00:00
        2018-10-28 01:00:00
        2018-10-28 02:00:00
        2018-10-28 03:00:00
        2018-10-28 04:00:00
        2018-10-28 05:00:00
]

In [15]: s.dt.replace_time_zone('Europe/Amsterdam', ambiguous='null')
Out[15]: 
shape: (6,)
Series: 'literal' [datetime[μs, Europe/Amsterdam]]
[
        2018-10-28 00:00:00 CEST
        2018-10-28 01:00:00 CEST
        null
        2018-10-28 03:00:00 CET
        2018-10-28 04:00:00 CET
        2018-10-28 05:00:00 CET
]

towards #11579. Based on #14842 (comment), I think there's a need for being able to just have null for problematic datetimes.

I'm suggesting to start with adding ambiguous='null', and then adding the option non_existent: Literal['null', 'raise']

Perf implication:

• for the trivial case of setting/unsetting UTC and ambiguous='raise' (the default), no implication. There's already a fast-path for that
• base case (ambiguous='raise', the default): no impact. ambiguous: no impact. I've made a fastpath to preserve this case. Check there's no impact here, which shows 12.373298853000051 => 12.362355302333375 (main => here)
• For the non-base-case, I'm seeing a slight perf hit, which I've timed here: 13.127492245999747 => 13.452795328000017 (main => here). I think this is to be expected, as the result needs to be constructed from a vector of PolarsResult<Option<i64>> as opposed to PolarsResult<i64>, but I think it's worth it for the user benefit

[codecov]

Codecov Report

Attention: Patch coverage is 94.59459% with 6 lines in your changes are missing coverage. Please review.

Project coverage is 80.99%. Comparing base (8c2eca9) to head (37cfb09).

❗ Current head 37cfb09 differs from pull request most recent head cce0c4c. Consider uploading reports for the commit cce0c4c to get more accurate results

Files	Patch %	Lines
crates/polars-arrow/src/legacy/kernels/time.rs	75.00%	3 Missing ⚠️
...ps/src/chunked_array/datetime/replace_time_zone.rs	96.10%	3 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main   #14961   +/-   ##
=======================================
  Coverage   80.99%   80.99%           
=======================================
  Files        1333     1333           
  Lines      173174   173229   +55     
  Branches     2458     2458           
=======================================
+ Hits       140265   140312   +47     
- Misses      32442    32449    +7     
- Partials      467      468    +1     

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

[stinodego]

I'll leave the Rust review to someone else, but I'm happy with the addition of the "null" option!

[MarcoGorelli]

separately, a non_existent argument could be added to make sure that this would return null values instead of raising. This is just to check that ambiguous='null' doesn't affect what happens for non-existent datetimes (coverage purposes)

EDIT: #15062 opens the doors for non_existent