feat: Add cum_count expression function

[stinodego]

Ref #13473
Ref #12420

When used without arguments, this is basically just syntactic sugar for int_range(1, count()+1, 1, dtype=IndexDtype).
I added this to the Rust side as we have no notion of IndexType in Python and I think it's nice to have in Rust anyway.

With arguments it's syntactic sugar for pl.col(...).cum_count(). It makes sense to have this along the other 'shortcut' functions we already have.

[cmdlineluser]

Perhaps it could also be a good time to deprecate the reverse= named arg.

pl.cum_count(reverse=True)
pl.cum_count().reverse()

(Expr.reverse() did not exist when .cumcount() was initially added - hence the need for the named arg at the time.)

[stinodego]

Perhaps it could also be a good time to deprecate the reverse= named arg.

I'm not entirely sure. All cum_* functions have a reverse option, might as well include it in this one.

While it is technically redundant, otherwise you have to write df.select(pl.col("a").reverse().cum_sum().reverse()). Which feels bad.
For the cum_count function, a single reverse will do the trick. But let's be consistent and keep the reverse param for all cumulative functions.

[rben01]

This counts nulls as +1? Seems confusing when df.count() skips over nulls; I would expect the last element in the column represented by pl.col("x").cum_count() to be the same as df.count().get_column("x"). Like this:

┌──────┬───────────┬─────────┐
│ a    ┆ cum_count ┆ cum_len │
│ ---  ┆ ---       ┆ ---     │
│ i64  ┆ i64       ┆ i64     │
╞══════╪═══════════╪═════════╡
│ 1    ┆ 1         ┆ 1       │
│ null ┆ 1         ┆ 2       │
│ 3    ┆ 2         ┆ 3       │
└──────┴───────────┴─────────┘

Not sure if cum_len is a good name — other options include numbered, row_number, row_index1 (or just skip it altogether and use expr.with_row_index() + 1) — but I would definitely consider cum_count misleading.

[stinodego]

I don't know what you mean. It functions exactly as you say it does - we made it match df.count(). Both ignore null values now.

import polars as pl

df = pl.DataFrame({"a": [1, 2, 3], "b": [4, None, 6]})

print(df.count())
print(df.select(pl.col("a", "b").cum_count()))

shape: (1, 2)
┌─────┬─────┐
│ a   ┆ b   │
│ --- ┆ --- │
│ u32 ┆ u32 │
╞═════╪═════╡
│ 3   ┆ 2   │
└─────┴─────┘
shape: (3, 2)
┌─────┬─────┐
│ a   ┆ b   │
│ --- ┆ --- │
│ u32 ┆ u32 │
╞═════╪═════╡
│ 1   ┆ 1   │
│ 2   ┆ 1   │
│ 3   ┆ 2   │
└─────┴─────┘

[rben01]

Oh, sorry never mind.. I think I got confused by the int_range(1, count()+1, 1, dtype=IndexDtype) part unconditionally returning row numbers. (And by the fact that elsewhere in polars, such as group_by, count means number of rows, not number of non-nulls.)