allow fiberassign --stdstar duplicates with --mtl

[sbailey]

This PR fixes #198 by allowing fiberassign to have duplicate targets in the --stdstar file and the --mtl. It also exposures the fba_run options --sciencemask, --stdmask, etc. to fiberassign.

Background: Legacy fiberassign allowed duplicate std/mtl targets and it worked due to how it did internal bookkeeping. The refactored fba_run does not allow duplicate targets, but that broke some of the ways we had been using it.

This PR: If all --stdstar targets are already in MTL, it just ignores the file. If it has some unique targets, those are filtered out and written to a separate temporary file before passing on to fba_run. This allows users to continue to use fiberassign in the same manner as before, without worrying about std/mtl duplicates.

[sbailey]

Hmm; a bunch of Travis test failures, but tests pass at NERSC. Newer fitsio is first suspicion since it made some non-backwards compatible changes.

[sbailey]

fitsio 1.0.x switched to reading string table columns as unicode strings instead of raw bytes, requiring tests like

    science_rows = np.where(fiber_data["DEVICE_TYPE"] == b"POS")[0]

to be changed to

    science_rows = np.where(fiber_data["DEVICE_TYPE"].astype(str) == "POS")[0]

to work with both 0.9.x and 1.0.x

Now I'm getting intermittent unicode parsing crashes from fitsio, e.g.

======================================================================
ERROR: test_io (fiberassign.test.test_assign.TestAssign)
----------------------------------------------------------------------
UnicodeDecodeError: 'ascii' codec can't decode byte 0xcb in position 0: ordinal not in range(128)

The above exception was the direct cause of the following exception:

SystemError: <class 'UnicodeDecodeError'> returned a result with an error set

[repeated many times]

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/home/travis/build/desihub/fiberassign/py/fiberassign/test/test_assign.py", line 154, in test_io
    read_assignment_fits_tile((tid, infile))
  File "/home/travis/build/desihub/fiberassign/py/fiberassign/assign.py", line 621, in read_assignment_fits_tile
    fbassign = fd["FIBERASSIGN"].read()
  File "/home/travis/virtualenv/python3.6.3/lib/python3.6/site-packages/fitsio/hdu/table.py", line 604, in read
    data = self._read_all(**keys)
  File "/home/travis/virtualenv/python3.6.3/lib/python3.6/site-packages/fitsio/hdu/table.py", line 649, in _read_all
    array = self._maybe_decode_fits_ascii_strings_to_unicode_py3(array)
  File "/home/travis/virtualenv/python3.6.3/lib/python3.6/site-packages/fitsio/hdu/table.py", line 1341, in _maybe_decode_fits_ascii_strings_to_unicode_py3
    array = array.astype(new_dt, copy=False)
SystemError: <class 'UnicodeDecodeError'> returned a result with an error set

I'm now trying to catch it in the act and cache the problematic test file to see if it is a problem with fitsio or with something we're writing. But it doesn't fail every time the tests run...

[sbailey]

Unicode parsing errors are our fault. assign.py line 859 creates an empty table to fill with fiber assignments:

    tile_targets = np.empty(len(tile_tgids), dtype=tile_targets_dtype)

That includes a BRICKNAME string column that never gets initialized, and thus randomly sometimes contains non-ASCII characters which aren't allowed in FITS.

For this particular case, I think replacing it with np.zeros and also actually filling in the brickname would fix this problem. However, there are multiple other uses of np.empty throughout fiberassign, so pausing for a moment to consult with @tskisner . I'd prefer to take the minor performance hit and initialize these other cases with np.full(length, bogus_value, dtype), and then include checks at the end of initialization that no bogus values remain. Thoughts?

[sbailey]

FYI I'm proceeding with my proposed replacement of np.empty -> np.zeros and np.full.

[sbailey]

For the record: I handled the uninitialized variables case, but I didn't add the (slow) code to fill in BRICKNAME in this PR.