Should we allow the reference coordinate and pixel size of
output image to be defined by input parameters?
Current implementation of the re-projection is a little
inefficient.
Current implementation loads all image data into memeory
simultaneously. This could cause problems if mosaicing a large number
of images.
The task currently assumes that exposure images are listed in
the same order as their raw photon twin, and the WCS keywords of
exposure images are taken from the raw image.
Task should be capable of writing output image as any (sensible)
data type, chosen by input parameter.
Currently only point-sampling is supported.
Limits should be placed on the size of the output image.
