ENH: Add .content, .text and .json properties to NIfTI extensions

[effigies]

The NIfTI-MRS standard (https://github.com/wtclarke/mrs_nifti_standard/) includes a JSON extension. My initial attempt at this (in #1327) created a NiftiJSONExtension class, but this had the side-effect of breaking code that already loaded the extension as bytes and manually decoding.

This PR instead includes the following protocol:

class NiftiExtension:
    content: bytes
    text: str
    json: dict[str, Any]

This follows the protocol used by requests and httpx and assumes that most extensions will be sufficiently served by one or another of these three attributes (properties, actually), and mangling/unmangling is only needed for more intricate types, such as DICOM.

Nifti1Extension is a non-ideal base class for NIfTI extensions because it assumes that it is safe to use an identity transformation to convert between the on disk and in memory representation of the extension contents, and thus default to bytes objects. This makes it difficult to define its typing behavior in a way that allows subclasses to refine the type such that type-checkers understand it.

This patch creates a generic NiftiExtension class that parameterizes the "runtime representation" type. Nifti1Extension subclasses with another parameter that defaults to bytes, allowing it to be subclassed in turn (preserving the Nifti1Extension -> Nifti1DicomExtension subclass relationship) while still emitting bytes. Further, this allows extensions to be created with either bytes or an object (for subclasses that define a particular object type), avoiding unnecessary mangle/unmangle round-trips.

We could have simply made Nifti1Extension the base class, but the mangle/unmangle methods need some casts or ignore comments to type-check cleanly. This separation allows us to have a clean base class with the legacy hacks cordoned off into an subclass.

The Cifti2Extension needed very little updating.

Closes #1335.

[codecov]

Codecov Report

Attention: Patch coverage is 96.42857% with 4 lines in your changes missing coverage. Please review.

Project coverage is 95.34%. Comparing base (83eaf0b) to head (398488e).

Files with missing lines	Patch %	Lines
nibabel/nifti1.py	95.45%	3 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1336      +/-   ##
==========================================
- Coverage   95.34%   95.34%   -0.01%     
==========================================
  Files         207      207              
  Lines       29507    29555      +48     
  Branches     4982     4993      +11     
==========================================
+ Hits        28134    28179      +45     
- Misses        932      935       +3     
  Partials      441      441              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[effigies]

A review would be appreciated here. I got back to this because I wanted to play around with serializing xibabel images to NIfTI, optionally dumping "unknown" metadata fields into a JSON extension, and this will make that more convenient.

Does this seem like a reasonable interface?