Add MCDA module for spatial multi-criteria decision analysis (#1030)

[brendancol]

Closes #1030

Summary

• New xrspatial.mcda subpackage with five modules: standardize, weights, combine, constrain, sensitivity
• Six standardization methods (linear, sigmoidal, gaussian, triangular, piecewise, categorical) for converting raw criterion layers to a 0-1 suitability scale
• AHP pairwise comparison weights with consistency ratio validation, plus rank-order weighting (ROC, rank sum, reciprocal)
• Five combination methods: WLC, WPM, OWA (tunable risk attitude), fuzzy overlay (AND/OR/sum/product/gamma), boolean overlay
• Constraint masking and sensitivity analysis (one-at-a-time perturbation and Monte Carlo CV)
• All operations are element-wise, so they work across numpy, cupy, dask+numpy, and dask+cupy without explicit backend dispatch
• No new dependencies

Test plan

• 148 tests passing (pytest xrspatial/tests/test_mcda.py)
• Correctness against known values (AHP textbook example, ROC formula, WLC/WPM manual computation)
• Edge cases: degenerate triangles, inf/NaN propagation through all methods, single-criterion datasets, single-pixel rasters, sigmoid overflow, inverted bounds, duplicate breakpoints, AHP self-comparison/zero/negative, extra weight keys, boolean overlay with floats, mismatched coords
• Dask: chunk alignment for standardize/WLC/fuzzy/OAT, Monte Carlo eager computation to prevent graph explosion, piecewise/categorical map_blocks cupy safety
• Sensitivity: weight boundary clamping, single-criterion no-op, Monte Carlo reproducibility (same seed = same result), WPM combine method
• Docs reference page and README feature matrix updated
• User guide notebook (35_MCDA.ipynb)