feat: Add new Enum categorical data type which allows a fixed set of categories

[c-peters]

Relates to #10705

This PR is meant as a start of a sequence of PRs to improve the categoricals in Polars
This allows users to provide a fixed list of categories when initializing / casting to a categorical

s = pl.Series('a', ['a','b','c'], dtype=pl.Enum(['a','b','c']))

df = pl.LazyFrame({'a':['a','b','c']}).with_columns(pl.col('a').cast(pl.Enum(['a','b','c']))).collect()

Specifying a value outside of the provided list will create an error

s = pl.Series('a', ['a','b','c'], dtype=pl.Enum(['a','b']))

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/chiel/Documents/polars/polars/py-polars/polars/series/series.py", line 283, in __init__
    self._s = sequence_to_pyseries(
  File "/home/chiel/Documents/polars/polars/py-polars/polars/utils/_construction.py", line 423, in sequence_to_pyseries
    pyseries = pyseries.cast(dtype, strict=True)
exceptions.OutOfBoundsError: Value c in string column not found in fixed set of categories LargeUtf8Array[a, b]

Note that this is in addition to the categorical. The distinction is that categorical types are flexible and new categories get added on the fly while with Enum they are fixed and don't change.

Todo in future PRs

• Allow casting from Uint32 as the index into fixed set of categoricals (see Allow creating Categorical with pre-defined categories #10705)
• Think about ordering on the provided list
• Allow operations (e.g. joins) on the categorical with equal categories without the global string cache

[stinodego]

@c-peters looks like you have to rebase on main to get the lints to pass! looks like you already did 😅

[ritchie46]

Hope to get to this one end of today.

[stinodego]

I know this is a WIP but I spotted some minor things on the Python side - figured I might as well leave a comment.

[c-peters]

Waiting on #12091 to fix failing test that are due to another bug

[stinodego]

This looks great! It's going to make a lot of users very happy.

I left a whole bunch of nitpick comments but overall this looks solid, at least on the Python side (I'll leave the Rust side to Ritchie for now).

We have some ways to go for better integration of this data type such as making sure it works with the interchange protocol __dataframe__ method and probably some others, but we can pick those up as we go along.

[ritchie46]

Can you rebase? Then we can get this in!

[stinodego]

Lint failure is my fault (dependency updates) - fixing as we speak. Another rebase is probably needed.

EDIT: Rebased.