MsgPack bug with NumPy 1.26

[jakirkham]

From @h-vetinari in conda-forge/numcodecs-feedstock#96 (comment) :

This looks like an issue with using deprecated numpy functionality that got removed in 1.25 or 1.26:

self = MsgPack(raw=False, use_bin_type=True, use_single_float=False)
buf = [[0], [1], [2, 3]]

    def encode(self, buf):
>       buf = np.asarray(buf)
E       ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (3,) + inhomogeneous part.

$PREFIX/lib/python3.10/site-packages/numcodecs/msgpacks.py:55: ValueError
=========================== short test summary info ============================
FAILED tests/test_json.py::test_non_numpy_inputs - ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (3,) + inhomogeneous part.
FAILED tests/test_msgpacks.py::test_non_numpy_inputs - ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (3,) + inhomogeneous part.
============ 2 failed, 536 passed, 34 skipped, 38 xfailed in 14.91s ============

[jakirkham]

Nvm. Looks like this was already fixed with PR: #417