Accept numpy integers as train_size / test_size in train_test_split

[LeSingh1]

Dataset.train_test_split() uses isinstance(..., int) / isinstance(..., float) to tell "a number of rows" from "a fraction", and rejects everything else:

if train_size is not None and not isinstance(train_size, (int, float)):
    raise ValueError(f"Invalid value for train_size: {train_size} of type {type(train_size)}")

numpy floats are subclasses of float, so they already work. numpy integers are not subclasses of int, so they don't:

import numpy as np

ds.train_test_split(test_size=np.float64(0.2))   # works
ds.train_test_split(test_size=np.int64(2))
# ValueError: Invalid value for test_size: 2 of type <class 'numpy.int64'>

The asymmetry is accidental, and the error is confusing because the printed value (2) looks perfectly valid. numpy integers turn up naturally — np.floor(0.2 * len(ds)).astype(int), a value read out of a numpy array, a pandas .nunique(), etc. sklearn.model_selection.train_test_split, which this API is modelled on, accepts them.

This PR converts numpy integers to int before the checks, leaving the int/float semantics untouched.

Added tests/test_arrow_dataset.py::test_train_test_split_with_numpy_integer_sizes, parametrized over int32, int64 and uint8, covering both test_size and train_size. All three fail on main.