feat(rust): guarantee schema-stable col(dtype) selection

[alexander-beedie]

Issue

• Using dtype-based column selection with more than one dtype (eg: polars.datatypes.NUMERIC_DTYPES) was not guaranteed to be stable between sessions/platforms/etc, as those dtype groups are set-based.
• Other (non-set) multi-dtype selections return columns in dtype-order, not schema-order.

Solution

Switch the inner/outer filter-order in expand_dtypes when determining which columns match.
Guarantees stable column selection while maintaining the underlying schema-order.

Example

from polars.datatypes import NUMERIC_DTYPES
import polars as pl

df = pl.DataFrame(
    data=[],
    schema={
        "a": pl.Int8,
        "b": pl.Int16,
        "c": pl.Int32,
        "d": pl.Int64,
        "e": pl.Float32,
        "f": pl.Float64,
        "g": pl.UInt8,
        "h": pl.UInt16,
        "i": pl.UInt32,
        "j": pl.UInt64,
    },
)

Before: (unstable column order - changes between sessions)

df.select( pl.col(NUMERIC_DTYPES) )
# ┌─────┬─────┬─────┬─────┬─────┬─────┬─────┬─────┬─────┬─────┐
# │ f   ┆ c   ┆ e   ┆ b   ┆ j   ┆ a   ┆ i   ┆ h   ┆ g   ┆ d   │
# │ --- ┆ --- ┆ --- ┆ --- ┆ --- ┆ --- ┆ --- ┆ --- ┆ --- ┆ --- │
# │ f64 ┆ i32 ┆ f32 ┆ i16 ┆ u64 ┆ i8  ┆ u32 ┆ u16 ┆ u8  ┆ i64 │
# ╞═════╪═════╪═════╪═════╪═════╪═════╪═════╪═════╪═════╪═════╡
# └─────┴─────┴─────┴─────┴─────┴─────┴─────┴─────┴─────┴─────┘

After: (stable, and schema-ordered)

df.select( pl.col(NUMERIC_DTYPES) )
# ┌─────┬─────┬─────┬─────┬─────┬─────┬─────┬─────┬─────┬─────┐
# │ a   ┆ b   ┆ c   ┆ d   ┆ e   ┆ f   ┆ g   ┆ h   ┆ i   ┆ j   │
# │ --- ┆ --- ┆ --- ┆ --- ┆ --- ┆ --- ┆ --- ┆ --- ┆ --- ┆ --- │
# │ i8  ┆ i16 ┆ i32 ┆ i64 ┆ f32 ┆ f64 ┆ u8  ┆ u16 ┆ u32 ┆ u64 │
# ╞═════╪═════╪═════╪═════╪═════╪═════╪═════╪═════╪═════╪═════╡
# └─────┴─────┴─────┴─────┴─────┴─────┴─────┴─────┴─────┴─────┘

Note

Selection in dtype order remains trivial (and each set of matching cols is also now stable), eg:

df.select( [pl.col(FLOAT_DTYPES), pl.col(INTEGER_DTYPES), ...] )