Implement the full loop for MTL

[geordie666]

This PR implements the full loop of constructing a zcat by reading "done" tiles from a specified directory, determining which "done" tiles are yet to be processed, updating ledgers, and creating and updating an "MTL tile file" to record which tiles have been processed through MTL. New features include:

• Modify the ledgers based on any new (i.e. previously unprocessed) tiles in a zcat directory.
• Create and update an "MTL tile file" which tracks which tiles have been processed by MTL.
• Read the "standard" tile file to determine the observing conditions for each tile, and only process a given tile for the appropriate observing conditions.
  • This should also only process tiles if they are in the appropriate tile file for a given survey flavor (sv2, main, etc.).
• An option to use the ledgers themselves to update NUMOBS instead of expecting NUMOBS to be in the zcat.
  • Currently this option is turned on, but it will be easy to turn off if we decide to track NUMOBS (e.g. via NUMTILE) in the zcat.
• A command-line script to execute the full loop. Each of the required input directories can be passed or can be environment variables.
  • An example is: run_mtl_loop DARK --survey sv2 --zcatdir $ZCAT_DIR --mtldir $MTL_DIR --tilefn $TILE_FN

A number of refinements are likely to be needed that build on this PR, as:

• We haven't yet converged on the data model and full directory path for the zbest files.
• Eleanor will eventually want to drop in a different make_zcat() function to replace my simple make_zcat_rr_backstop().
• We haven't yet determined whether we will use NUMTILE from the spectroscopic pipeline outputs, or if we will let the ledger update NUMOBS itself.
• We will need to update priorities and NUMOBS more carefully based on ZWARN

But this is likely enough infrastructure to have a first crack at the 0.1% survey.

[coveralls]

Coverage decreased (-0.5%) to 59.648% when pulling 71fd9a5 on ADM-MTL-loop into 4fc5e72 on master.