Add complex dtype support to FillValueCoder for Zarr backend

[maxrjones]

This PR adds support for encoding/decoding complex fill values (complex64, complex128) in the Zarr backend's FillValueCoder, which previously raised ValueError for complex dtypes. This enables converting/virtualizing NetCDF files with these fill values to Zarr. As an example, this change unlocks virtualizing full NISAR granules using VirtualiZarr + Xarray.

Each component (real/imag) is base64-encoded as a little-endian double, consistent with how float fill values are already handled and ensuring safe JSON round-tripping of special values like NaN and Infinity.

• Closes #xxxx
• Tests added
• User visible changes (including notable bug fixes) are documented in whats-new.rst
• New functions/methods are listed in api.rst

[maxrjones]

@dcherian thanks for your review. Do I need to get mypy working for this PR to go into main? The failures are in the function that I worked on, but unrelated to the changes that I made.

[jsignell]

Looks good! Mentioned offline that I think the suggestion in https://github.com/pydata/xarray/pull/11151/changes#r2805178961 was to write a hypothesis test, so just that and one little nit.

[jsignell]

@maxrjones the hypothesis test looks good. Whenever you are done, just add a line to "what's new" and then this is good to merge.

[maxrjones]

@maxrjones the hypothesis test looks good. Whenever you are done, just add a line to "what's new" and then this is good to merge.

thanks, Julia! I added this as a "feature", but wasn't 100% confident in the categorization. Please let me know if I should move it elsewhere.