fix isLRG vs. isLRG_colors

[sbailey]

This PR

• switches to using isLRG_colorcuts instead of isLRG when applying cuts to mock spectra (Similar to use isLRG_colors for color-cuts desisim#246)
• fixes random seed propagation when processing pixels in parallel (it was fixed for 2% sprint but I accidentally deleted my git repo before committing, so refixing now...)

Smaller but important changes to get things back into a working state:

• fix a vestigial (and crash inducing) clobber=True in mpi_select_mock_targets
• drop back to processing a single healpixel at a time per rank instead of groups of healpixels to avoid blowing out memory.
• don't impose DEBUG-level logging on everyone that imports desitarget.mock.io
  (see Questions/suggestions for logging module desiutil#66 for a longer-term cleanup of this type of problem)
• use x[col][:] = values instead of x[col] = values which changes the dtype starting with astropy 1.3, resulting in output files having different formats (e.g. S10 vs. S13) for different pixels.