Changes

Fix windowed reproject window construction
Investigating strange behavior in windowed reprojection, it was realized
that the gridbounds being used to construct a window was also being used
to construct the expected RasterExtent of said window. This is not
desirable because the unclamped window can be reasonably expected to be
smaller than the grid used to construct it. This entails more expected
cells (and therefore a smaller cell size) than is desirable. Cross
referencing our result with GDAL validated this concern. See:
https://github.com/locationtech/geotrellis/pull/3095