Add landscape fragmentation metrics and habitat connectivity

[brendancol]

Ecology and conservation workflows need landscape pattern metrics and connectivity analysis. Neither is currently available in xarray-spatial.

Scope

Landscape metrics (FRAGSTATS-style)

Patch-level and landscape-level metrics computed on categorical rasters:

• Patch area, perimeter, shape index
• Edge density
• Core area
• Fractal dimension
• Contagion

These should work both per-class and landscape-wide.

Fragmentation assessment

• Effective mesh size
• Patch cohesion

Resistance-surface connectivity

• Least-cost corridor networks between multiple source patches. This extends the existing pairwise least_cost_corridor() to handle N source patches simultaneously and produce a combined corridor surface.
• Circuit-theory connectivity (Circuitscape-style current flow). This requires a sparse linear solve and is computationally expensive, but it's in very high demand for conservation planning.

Design notes

Landscape metrics operate on categorical rasters and pair naturally with existing regions() for patch identification. The regions output can feed directly into metric computation.

Circuit-theory connectivity is the most complex piece here. A basic implementation could use scipy sparse solvers on CPU and cuSparse on GPU. Even a simplified version (e.g., pairwise resistance distance without full current maps) would be useful.