pRasterBlaster is a lot slower than gdalwarp in serial case

[kornholi]

Hao brought this up. 5s vs 200s even with no resampling.

Majority of pRB time is spent in GDAL/PROJ4 reprojecting points (proj4 TransformEx), so this is interesting. Maybe GDAL takes an invalid shortcut (that works fine in most cases)?

More investigation needs to be done.