FITS file size, and type

Add dtype='float32' to ccdproc functions, and skip the "manual" astype. This is cleaner, as it also affects the UNCERT HDU, which otherwise remains at 64 bit.

Indeed currently we have 32 bit PRIMARY, 8 bit MASK, 64 bit UNCERT.


We could maybe also try to avoid generating the MASK and UNCERT HDUs as we don't use them here, but that would be working against ccdproc.