Initial Zarr v3 Support

[tasansal]

Refactored the Zarr storage and array APIs to work with Zarr v3 engine. The datasets are still created as Zarr v2; however, I/O operations are now done with v3's new async engine.

Zarr v3 minimum Python req is 3.11 so I updated MDIO to that as well.

Good Stuff:

• Non-Dask reads are now also fast and no Dask overhead.
• We can switch to v3 when struct array support is done: [v3] Structured dtype support zarr-developers/zarr-python#2134

Limitations:

• No v3 sharding since arrays are v2
• There is a bug with v3 and struct arrays: [v3] Structured dtype support zarr-developers/zarr-python#2134 (comment). However we can still read them with slicing.
• copy_store is not implemented in v3 yet, can't copy with MDIO API.

TODO:

• Make tests work
• Validate if rechunk API is working
• Test accessor patterns (reader, writer, accessor)
• Maybe implement copy_store in https://github.com/zarr-developers/zarr-python
• Maybe implement struct arrays in https://github.com/zarr-developers/zarr-python

This branch is based on #496