General clean-ups and speed-ups for DR9 work

[geordie666]

This PR cleans-up (and speeds-up) some functionality as we move towards producing DR9 targets and fiberassign files. It includes:

• A corrected data model when repartitioning skies into their correct HEALPixels.
  • skies are run in bricks that are centered in HEALPixels, rather than in the actual boundaries of a HEALPixel, so they need to be "repartitioned" into the correct pixels (see Repartition sky files so sky locations lie within HEALPixel boundaries #621); this PR changes the organization of the input/output files to automatically reflect the data model that I have started to use when releasing targets.
• Faster versions of all of the desitarget.io.read_targets_in_X functions:
  • i.e., read_targets_in_tiles(quick=True), read_targets_in_box(quick=True), read_targets_in_cap(quick=True), read_targets_in_hp(quick=True).
  • The new functions are less general, but they run 2-10x faster.
  • The speed-up is achieved by assuming the HEALPixel-split data model rather than by checking and determining the data model from header information (so the "quick=True" versions are more fragile).
  • The speed-up is greater for HEALPixel-split files at higher nside. Comparing targets partitioned into files at nside=2 versus nside=8, the speed-up was 5x.
• Establishing "standard" MASKBITS combinations for all of the target classes that can be read via convenience functions, and using those convenience functions to automatically inform the "FRACAREA_X" calculations in pixweight files.
  • This makes it far easier to ensure that the MASKBITS cuts adopted by the target files are always aligned with those calculated in the pixweight files.
• Catching the case where a user tried to construct MTL ledgers at a lower resolution than the corresponding target files (which partially addresses Ledger code failure for high to low res, e.g. with mock targets. #650).