Accept numpy integers for rollout nstep and chunk_size

[Nas01010101]

Problem

rollout rejects nstep and chunk_size arguments whose type is a numpy integer (e.g. np.int64):

if nstep and not isinstance(nstep, int):
    raise ValueError('nstep must be an integer')
if chunk_size and not isinstance(chunk_size, int):
    raise ValueError('chunk_size must be an integer')

isinstance(np.int64(3), int) is False, so a numpy integer is rejected even though it is a valid integer. These values very commonly come from numpy operations (array.shape[i], products of dimensions, np.arange(...)[i], etc.), so this raises ValueError: nstep must be an integer:

rollout.rollout(model, data, initial_state, nstep=np.int64(3))

while the identical call with a Python int works.

Fix

Accept numpy.integer in addition to the built-in int for both checks (numpy is already imported as np). The downstream C++ call already handles the value numerically, so no other change is needed.

Tests

Adds test_numpy_integer_nstep_and_chunk_size, which calls rollout with np.int64 values for nstep and chunk_size. It fails before this change (ValueError) and passes after; the rest of rollout_test.py remains green.