Investigate making numpy dependency optional

[dstansby]

#112 is a proposal to make the zarr dependency optional. On the face of it, it looks like we could also make numpy optional too, but knowing for sure would probably require someone trying to make it optional and seeing if it would work.

[d-v-b]

because Zarr V2 uses a data type representation that's based on numpy, it's super useful to have "turn this numpy data type into a string" semantics in this library, and numpy is a great source for that kind of thing. See

pydantic-zarr/src/pydantic_zarr/core.py

Lines 55 to 86 in 0b9bd1d

	def parse_dtype_v2(value: npt.DTypeLike) -> str | list[tuple[Any, ...]]:
	"""
	Convert the input to a NumPy dtype and either return the ``str`` attribute of that
	object or, if the dtype is a structured dtype, return the fields of that dtype as a list
	of tuples.
	Parameters
	----------
	value : npt.DTypeLike
	A value that can be converted to a NumPy dtype.
	Returns
	-------
	A Zarr V2-compatible encoding of the dtype.
	References
	----------
	See the [Zarr V2 specification](https://zarr-specs.readthedocs.io/en/latest/v2/v2.0.html#data-type-encoding)
	for more details on this encoding of data types.
	"""
	# Assume that a non-string sequence represents a the Zarr V2 JSON form of a structured dtype.
	if isinstance(value, Sequence) and not isinstance(value, str):
	return [tuple(v) for v in value]
	else:
	np_dtype = np.dtype(value)
	if np_dtype.fields is not None:
	# This is a structured dtype, which must be converted to a list of tuples. Note that
	# this function recurses, because a structured dtype is parametrized by other dtypes.
	return [(k, parse_dtype_v2(v[0])) for k, v in np_dtype.fields.items()]
	else:
	return np_dtype.str

for an example.

I think we could get the same functionality without an explicit numpy dep by modelling numpy dtypes with a protocol. I don't think there are many false positives to worry about, so this should be safe.