fix: Improve literals for temporal subclasses

[mcrumiller]

Fixes #18906.

import pandas as pd
import polars as pl

pl.select(a=pl.lit(pd.Timedelta(minutes=1)))
# shape: (1, 1)
# ┌──────────────┐
# │ a            │
# │ ---          │
# │ duration[μs] │
# ╞══════════════╡
# │ 1m           │
# └──────────────┘

This update pushes allow_object into the AnyValue conversion, and by doing so avoids the double-gil lock for subclassed items (@ritchie46 I think it's what you were thinking in this comment). I think it should help tidy up some of the subclass issues with temporals. I also added checks on time and timedelta temporal classes as well, as is the case in the linked issue where pd.Timedelta subclasses datetime.timedelta.

[mcrumiller]

This was unused as far as I can tell. Removing it didn't fail any tests. The path to a literal list is elsewhere.

[codecov]

Codecov Report

Attention: Patch coverage is 98.41270% with 1 line in your changes missing coverage. Please review.

Project coverage is 79.88%. Comparing base (901b243) to head (a61f379).
Report is 2 commits behind head on main.

Files with missing lines	Patch %	Lines
crates/polars-python/src/conversion/any_value.rs	97.43%	1 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main   #18998   +/-   ##
=======================================
  Coverage   79.88%   79.88%           
=======================================
  Files        1524     1524           
  Lines      207634   207643    +9     
  Branches     2904     2904           
=======================================
+ Hits       165873   165882    +9     
  Misses      41214    41214           
  Partials      547      547           

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

[mcrumiller]

I'm unable to replicate the failing test, even after installing same Python (3.12.16). @MarcoGorelli any ideas?

Edit: rebased and it went away.

[ritchie46]

Thanks @mcrumiller. I think this is a good improvement as we now convert in one location.

[ritchie46]

@mcrumiller can you take a look. This test failed in the ci: https://github.com/pola-rs/polars/actions/runs/11093722576/job/30820360138?pr=18930

[mcrumiller]

Yikes, sorry. Yeah let me look.

[mcrumiller]

Must be something about installed timezone modules, I can't reproduce. I'll work on the setup to figure out what's going on.

[mcrumiller]

For some reason get_datetime is being called sometimes when a subclass of time and timedelta are used, resulting in an invalid access to the tzinfo property. @MarcoGorelli do you have any clue what could trigger this?

[mcrumiller]

New theory: same class name defined in same function in multiple threads are colliding, even with different inheritence definitions.

[mcrumiller]

@ritchie46 I've added PR #19012 which I think resolves the issue.