convolve_2d: missing _validate_raster on input

[brendancol]

Summary

convolve_2d() in xrspatial/convolution.py doesn't validate its input before handing it to the backend kernel. Pass a boolean, string, or complex-dtype DataArray and you get a numba TypingError or a cupy crash instead of a ValueError that says what went wrong.

Cat 6 (input validation) finding deferred from PR #1243 (issue #1241).

Reproducer

import numpy as np
import xarray as xr
from xrspatial.convolution import convolve_2d, circle_kernel

agg = xr.DataArray(np.ones((10, 10), dtype=bool))
kernel = circle_kernel(1, 1, 2)
convolve_2d(agg.data, kernel)  # numba TypingError, not ValueError

Expected

A ValueError from _validate_raster naming the function and the bad dtype. PR #1384 already taught _validate_raster to reject complex dtypes, so one call covers boolean, string, and complex.

Fix

Call _validate_raster(xr.DataArray(data), func_name='convolve_2d', ndim=2) at the top of convolve_2d(), before the kernel runs.

Severity

MEDIUM. No exploit path; just confusing errors at the API boundary.