Avoid memory copy in local store write

[tomwhite]

This removes an unnecessary memory copy when writing to a local file, bringing the memory usage in line with Zarr Python v2 for this case (see https://github.com./tomwhite/memray-array).

TODO:

• Add unit tests and/or doctests in docstrings
• Add docstrings and API docs for any new/modified user-facing classes and functions
• New/modified features documented in docs/user-guide/*.rst
• Changes documented as a new file in changes/
• GitHub Actions have all passed
• Test coverage is 100% (Codecov passes)

[d-v-b]

thanks! this looks great.

[dstansby]

As the pre-commit typing failure indicates, doesn't f.write() need bytes given instead of a numpy array here? Maybe it doesn't since the tests all seem to pass - if that's the case, could you add an appropriate # ignore comment to fix the mypy error, and add a comment in the code explaining why it's okay to hand a numpy array directly to write()?

[TomAugspurger]

write supports anything implementing the buffer protocol, which ndarrays do. Prior to 3.12 that wasn't available in Python (https://peps.python.org/pep-0688/).

Bumping the python version in our mypy config at

zarr-python/pyproject.toml

Line 352 in 9e8b50a

python_version = "3.11"

might let this work.

[dstansby]

SPEC 0 has support Python 3.11 for a little while yet, so I don't think we should bump the Python version in the config. If it's all fine in practice, then just an ignore comment to make mypy happy should be fine.

[TomAugspurger]

Also needs a release note: https://zarr.readthedocs.io/en/stable/developers/contributing.html#changelog

Looks good once those are addressed.

[TomAugspurger]

The codecov failure seems to come from increasing the number of lines in an uncovered branch. #2859 is already tracking that, so I'll go ahead and merge this.