Speed-ups and bug-fixes for I/O routines

[geordie666]

This PR fixes a bug when reading MTL ledgers. It also implements some speed-ups in the functions that read targets, which were suggested by @araichoor. To summarize:

• When we moved to the final MTL data model, we broke the code that read Main Survey ledgers created by version 1.0.0 of desitarget (see also PR #742). The quick-fix that was implemented to address this issue broke the reader for SV targets (see issue #749). This PR implements a less fragile fix that facilitates reading all types of MTL ledgers, including SV3 and Main Survey 1.0.0.
• This PR implements a faster version of io.read_targets_in_tiles() via a new io.read_targets_in_tiles_quick() function.
• This PR also speeds up mtl.inflate_ledger() using a new mtl.match_ledger_to_targets() function.

[geordie666]

@araichoor: I think you can switch your bespoke code in fba_launch_io.py over to just two lines of code using the following schema:

1. Read the targets using targets=io.read_targets_in_tiles(hpdirname, tiles=tiles, quick=True) or by directly using targets=io.read_targets_in_tiles_quick(hpdirname, tiles=tiles)
2. Match the targets and the mtl using the quick version of mtl.inflate_ledger(), which is mtl_with_target_columns = mtl.match_ledger_to_targets(mtl, targets).

You might also want to just try mtl_with_target_columns = mtl.inflate_ledger(mtl, targdir, quick=True) directly, as there doesn't appear to be a huge speed-up from switching to using io.read_targets_in_tiles_quick().

It would be great if you could try this set-up for some pass 0 tiles and check that results are identical and (hopefully) faster than the standard desitarget.io routines.

[coveralls]

Coverage decreased (-0.2%) to 58.516% when pulling 752dde7f870735906738ed642cbe2361ca1ef9d5 on ADM-speed-IO into 263d22ea11f9df428bf6e2e5208b6681c9719312 on master.