NoData mask bands: Incorrect RasterIO result with mode resampling

[dbaston]

(Pdb) band.GetMaskBand().ReadAsArray()
array([[  0, 255],
       [  0,   0]], dtype=uint8)
(Pdb) band.GetMaskBand().ReadAsArray(buf_xsize=1, buf_ysize=1, resample_alg=gdal.GRIORA_Mode)
array([[255]], dtype=uint8)

Originally posted by @dbaston in #12172 (comment)

Some discussion in the linked PR.

[dbaston]

I had been thinking of having GDALNoDataMaskBand read the requested window into a memory dataset at full resolution, reclassify the values to 0 or 255, and then read from the in-memory dataset to the appropriate buffer size. This requires reading the entire window into memory at full resolution, which may be undesirable.

Alternatively, I've been thinking for a while about how it would be useful to have a VRT that easily does reclassification. (I guess this is possible as a VRT expression, but I'm thinking of something specific to this purpose.) Supposing such a VRT existed, it seems straightforward to for GDALNoDataMaskBand to construct a VRT that reads from its parent band, reclassifying the NoData value as 0 and everything else as 255. We could then read from this band into the appropriate buffer size.