Drop the hdf5storage dependency (write-side breaking change)

[KenyaOtsuka]

Summary

Removes the hdf5storage dependency entirely. This follows up on #137, which
moved the read path off hdf5storage (onto an h5py reader) but deliberately
left the dense/sparse write paths on hdf5storage.savemat, keeping the
dependency and the NumPy < 2 pin on Python 3.8/3.9.

The guiding principle here:

• Read compatibility is preserved. Existing files written by
  hdf5storage / MATLAB v7.3 (and by older bdpy) still load.
• Write compatibility with MATLAB is intentionally dropped. New files are
  written as bdpy-native plain HDF5 and are not meant to be opened by
  MATLAB's load.

This is a deliberate write-side breaking change (see Compatibility below).

Why

hdf5storage broke under NumPy 2.0 (it referenced the removed np.unicode_;
see #106). Rather than reimplement a full MATLAB-v7.3-compatible writer just to
keep the on-disk format, we drop the dependency and write plain HDF5 at the
save sites. bdpy reads its own output back through the existing h5py reader.

Changes

• bdpy/dataform/_mat_v73.py → read-only legacy reader

  • Remove savemat() / write_dataset() (and from __all__); the module no
    longer writes anything.
  • Rewrite the module docstring to state it provides h5py-based reading
    support for MATLAB v7.3 / hdf5storage .mat files.
  • Improve read_cell(): the plain-matrix branch now routes through
    read_dataset(), so MATLAB-style transposed matrices carrying
    MATLAB_class are de-transposed before being split into rows. Read
    behavior for bdpy's own (plain) index/shape datasets is unchanged.

• Write plain HDF5 directly at the save sites

  • bdpy/dataform/sparse.py: save_array (dense), save_multiarrays, and
    SparseArray.save write datasets/groups directly with h5py
    (SparseArray.save inlines the former __save_h5py body; append-mode still
    preserves other top-level variables).
  • bdpy/dataform/features.py: save_feature writes the feat dataset
    directly with h5py.

• Drop the dependency

  • Remove hdf5storage from pyproject.toml, mypy.ini, README.md, and the
    legacy tests/env/*/Pipfiles; clean up a stale comment in datastore.py.

Compatibility

	Behavior
Reading old hdf5storage / MATLAB v7.3 / bdpy files	✅ Unchanged
Newly written dense arrays	Plain HDF5 dataset
Newly written sparse arrays	Plain HDF5 group/datasets
Newly written feature files	Plain HDF5 with a feat dataset
Opening newly written files in MATLAB	❌ No longer supported (intentional)

The only compatibility promise is reading existing files. If MATLAB-readable
output is required, pin an older bdpy release.

Tests

• _mat_v73: removed the writer tests; added read_cell coverage for the
  transposed-matrix path.
• sparse: dense/sparse round-trips, save_multiarrays, preserve-other-
  variables, and an assertion that dense saves carry no MATLAB_class /
  Python.Shape.
• features: mock data now written as plain HDF5; added an explicit
  save_feature() test (round-trips via loadmat_key and Features).

[KenyaOtsuka]

This PR is intended to be merged in a release several releases after #142 has been merged.