Add bilateral filter for feature-preserving smoothing (#969)

[brendancol]

Summary

• Adds bilateral() for feature-preserving raster smoothing. Weights neighbors by spatial distance and value similarity, so noise gets smoothed but edges stay sharp.
• Four backends: NumPy (@ngjit), CuPy (@cuda.jit), Dask+NumPy, Dask+CuPy.
• Boundary modes (nan, nearest, reflect, wrap) and 3D multi-band arrays.

Closes #969.

What changed

File	Change
xrspatial/bilateral.py	New module with bilateral filter
xrspatial/__init__.py	Export bilateral
xrspatial/accessor.py	.xrs.bilateral() on DataArray and Dataset
xrspatial/tests/test_bilateral.py	18 tests: correctness, edge cases, cross-backend, boundary modes
docs/source/reference/focal.rst	API reference entry
examples/user_guide/17_Bilateral_Filter.ipynb	User guide notebook
README.md	Feature matrix row

Test plan

• Flat surface unchanged
• Sharp edges preserved with small sigma_range
• Edges blurred with large sigma_range
• NaN center pixels stay NaN, NaN neighbors skipped
• All-NaN and single-cell rasters handled
• Known-value correctness check
• Symmetry preserved
• Input validation rejects bad arguments
• NumPy, Dask+NumPy, CuPy, Dask+CuPy backends match
• Four boundary modes produce correct output
• 3D multi-band arrays processed per-band
• Accessor works