Add missing morphology accessor methods

[brendancol]

Closes #3399

The .xrs accessor exposed only four of the seven public morphology functions. This adds the missing three to both accessor classes.

• Add morph_gradient, morph_white_tophat, and morph_black_tophat methods to XrsSpatialDataArrayAccessor and XrsSpatialDatasetAccessor, following the existing forwarding pattern.
• Add accessor tests for the morphology family (none existed before) and extend the expected-methods checks to guard all seven methods.

The change forwards to the existing morphology functions, so backend coverage matches those: numpy, cupy, dask+numpy, dask+cupy.

Test plan:

• pytest xrspatial/tests/test_accessor.py (69 passed)
• New tests assert accessor output is identical to the direct function call for erode, gradient, white_tophat, black_tophat (DataArray) and gradient (Dataset)

[brendancol]

PR Review: Add missing morphology accessor methods

Blockers

None.

Suggestions

None.

Nits

None.

What looks good

• The three new methods follow the existing forwarding pattern byte-for-byte (lazy import, **kwargs passthrough), so they inherit whatever backend support the underlying functions provide.
• Both accessor classes are updated, so DataArray and Dataset stay in sync.
• Tests assert the accessor output is identical to the direct function call, which is the right contract for a thin forwarder.
• The expected-methods checks now list all seven morph methods, so a future regression that drops one gets caught.

Checklist

• Forwards to existing, already-tested morphology functions; no new algorithm
• Backend support inherited from the underlying functions (numpy / cupy / dask+numpy / dask+cupy)
• NaN handling unchanged (delegated)
• Accessor parity tested for DataArray and Dataset
• No materialization or copies introduced
• No benchmark needed (no compute path added)
• No README/docs change needed (accessor methods are not separately documented)
• Matches API conventions (mirrors the four existing morph accessor methods)