hotspots() skips kernel validation: bad kernels raise AttributeError or divide by zero

[brendancol]

Describe the bug

hotspots() in xrspatial/focal.py skips the kernel validation that apply() and focal_stats() both run. Those two call custom_kernel(kernel) before touching the kernel. hotspots() doesn't. It goes straight to _check_kernel_vs_raster_memory(kernel, ...) near line 1509, which reads kernel.shape directly.

So an invalid kernel either crashes with a confusing AttributeError or silently produces wrong output, depending on what you pass.

Failure modes (numpy backend)

• hotspots(agg, kernel=None) raises AttributeError: 'NoneType' object has no attribute 'shape' instead of a clear validation error.
• hotspots(agg, kernel=[[1, 1, 1]]) raises AttributeError: 'list' object has no attribute 'shape'. A list should get rejected with the same "not a Numpy array" message custom_kernel raises.
• hotspots(agg, kernel=np.ones((2, 2))) succeeds, even though even-dimensioned kernels are rejected by custom_kernel everywhere else.
• A zero-sum kernel (e.g. a Laplacian) divides by zero at line 1269: convolve_2d(data, kernel / kernel.sum(), boundary). It emits RuntimeWarnings and returns all-zero output instead of raising.

Expected behavior

hotspots() should validate its kernel the way apply() and focal_stats() do: route through custom_kernel() so None, lists, and even-dimensioned kernels raise clear, consistent ValueErrors. Zero-sum kernels should raise an explicit error rather than dividing by zero.

Additional context

custom_kernel() lives in xrspatial/convolution.py. It checks ndarray type and odd dimensions, but not the kernel sum. The zero-sum case is specific to hotspots() since it's the only focal function that divides by kernel.sum().