Fill remaining dask+cupy gaps: terrain, perlin, and zonal.crosstab

[brendancol]

Problem

Three functions each support 3 backends but are missing dask+cupy (and crosstab also missing cupy):

Function	NumPy	Dask	CuPy	Dask+CuPy
generate_terrain()	✅	✅	✅	❌
perlin()	✅	✅	✅	❌
crosstab()	✅	✅	❌	❌

References:

• xrspatial/terrain.py line 265: dask_cupy_func=not_implemented_func
• xrspatial/perlin.py line 247: dask_cupy_func=not_implemented_func
• xrspatial/zonal.py lines 1176–1180: both cupy_func and dask_cupy_func are not_implemented_func

Proposed Fix

• terrain dask+cupy: Wrap existing _terrain_cupy() in map_blocks (terrain generation is per-chunk independent).
• perlin dask+cupy: Wrap existing _perlin_gpu() in map_blocks.
• crosstab cupy: Adapt the numpy sort-and-stride counting logic using cupy sorting (similar to _stats_cupy pattern).
• crosstab dask+cupy: Combine cupy chunk function with dask delayed aggregation.

Impact

These gap-fillings complete the 4-backend matrix for more functions. While individually lower-priority than core analytics, they matter for users building end-to-end GPU pipelines — a pipeline that works on cupy but fails when wrapped in dask is frustrating.