First matmul implementation

[ricardosp4]

First implementation of matmul function along with its tests and the documentation. The tests should pass, as they pass locally (Windows).

Incidentally, I proposed a change in the squeeze method to better adapt to the array API.

[FrancescAlted]

Good job! See my comments.

[FrancescAlted]

Instead of using x1.dtype as the outcome, it would be better to use out_dtype = np.result_type(x1, x2).

[FrancescAlted]

Also, you should pass **kwargs to blosc2.zeros() for passing NDArray compression/storage details to the output.

[ricardosp4]

The exception error was this one:
> E numpy._core._exceptions._UFuncOutputCastingError: Cannot cast ufunc 'add' output from dtype('complex128') to dtype('float64') with casting rule 'same_kind'

What if that line is receives these arguments:
result = blosc2.zeros((n, m), dtype=np.result_type(x1, x2), **kwargs)

[FrancescAlted]

Yes, I was suggesting exactly that :-)

[FrancescAlted]

It would be nice to add tests that combines operands with different dtypes. Also, add tests that set kwargs and check that they work (by checking e.g out.cparams.codec).

[FrancescAlted]

I suggest to add a new test (say test_matmul_disk) where you pass filenames to operands and result via the urlpath keyword argument. See for example: https://github.com/Blosc/python-blosc2/blob/main/tests/ndarray/test_ndarray.py#L74

[ricardosp4]

I’ve added the test_matmul_disk as you suggested, passing the filenames through the urlpath argument. It's ready for review in commit 5e27651.

[FrancescAlted]

Looks good to me!