feat: Add list.index_of_in() to Expr and Series

[itamarst]

Fixes #20626.

The prototype in the draft PR (#20733) was 10× faster than using non-obvious hacky Python implementations based on list.eval(), so I would guess this is too.

Just as a reminder, the original requested feature in #5503 that led to the addition of Expr.index_of() was actually what's covered in this PR. And this is also what G-Research internal users had requested, so it seems like there are multiple potential users.

[codecov]

Codecov Report

Attention: Patch coverage is 97.50000% with 4 lines in your changes missing coverage. Please review.

Project coverage is 79.90%. Comparing base (635a215) to head (d6fa1d4).
Report is 63 commits behind head on main.

Files with missing lines	Patch %	Lines
...ars-plan/src/plans/conversion/type_coercion/mod.rs	80.00%	4 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #21192      +/-   ##
==========================================
+ Coverage   79.20%   79.90%   +0.70%     
==========================================
  Files        1585     1594       +9     
  Lines      225619   227899    +2280     
  Branches     2588     2600      +12     
==========================================
+ Hits       178697   182101    +3404     
+ Misses      46332    45201    -1131     
- Partials      590      597       +7     

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

[ritchie46]

We still haven't decided if we want this. I want to think of better ways than implementing every expression again over arrays and lists. Iterators over anyvalues, are things we should avoid. I want something more generic that allows us to reuse expressions on lists in way that doesn't impact binary size and reuses existing logic such as the row-encoding.

I first want to thing about this one. Being faster than list.eval isn't a metric we think is important in this, as list.eval is very very slow. This is of course perfectly implementable as plugin, if needed upstream.

[itamarst]

Insofar as possible this uses ChunkedArray.iter(), rather than Series.iter().

And the fact that (if you want type coercion in IR, anyway) it required an additional kind of type coercion is at least something to consider for a more generic implementation strategy.

If you decide not to accept this I will likely refactor into a plugin, yes.

[itamarst]

Not sure what's up with the Python docs.