Preserve coordinate dtypes

[nvictus]

Changes overlap to preserve the underlying integer type of the coordinate columns when casting output to nullable pandas dtypes.

EDIT: Changed to least surprising behavior:

1. Coordinate (start and end) columns in the input dataframes preserve dtypes in the output dataframe, when possible.
2. Outer join operations that introduce null values into coordinate columns will exhibit the native casting behavior of the column dtype: i.e. numpy columns get cast to float64 with NaNs while nullable pandas arrays remain unchanged (e.g. UInt32 -> UInt32) and contain NAs.

If one is worried about input-dependent casting behavior, one can convert coord columns to nullable types before passing to overlap with the.convert_dtypes() method. float64 output columns can also be forced into nullable Int64 after overlap by applying .convert_dtypes() on the output.

[gfudenberg]

sounds nice! should we add something like the following to tests?

df1.astype({"start": pd.Int64Dtype(), "end": pd.Int64Dtype()})
df2.astype({"start": uint32 "end": uint32})

overlap_dtypes = bioframe.overlap(df1, df2).dtypes
assert (df1.dtypes = overlap_dtypes[:3]).all()
assert (df2.dtypes = overlap_dtypes[3:]).all()

[nvictus]

Added tests.

[gfudenberg]

nice dtype test!

The only question I have is if the dtype conversions should go into a
Notes ---- section of the overlap docstring

[nvictus]

Added the note, and introduced an ensure_nullable option to force upcast to nullable dtypes in the input, which ensures the output coord columns are nullable (+ tests). Saves a few keystrokes from having to do it manually before applying overlap.