TARGETID=-1 is a useful statistics

[londumas]

Currently, in cell 57, the minitest rejects TARGETID=-1 objects:

# Merge the truth and zcat tables
# match_truth_z fails on the repeated TARGETID=-1 of unassigned fibers, so filter those out
assigned = zcat['TARGETID'] >= 0
dsq_z.match_truth_z(truth, zcat[assigned], mini_read=True)

However, these constitutes a very useful statistics, they work as a null-test:
In /project/projectdirs/desi/datachallenge/reference_runs/18.7/
there are 15 such objects, and 1 has a ZWARN==0,
the other have ZWARN=2**2=SMALL_DELTA_CHI2.

There is two issues:

• it would be nice to keep the statistics on these objects in the minitest.
  Maybe simply adding them in the truth table as UNPLUGGED with z=0
• The fact that they all have the same TARGETID=-1 causes problem when merging
  all the different zbest files together, and will not allow to keep track of all of them.
  In eBOSS we wouldn't have such an issue because objects have an id related to their
  observation: PLATE-MJD-FIBERID, not only their targeting.
  Would NUMEXP and NUMTILE allow such an id?

The failing object can be observed with the following command within ipython --pylab:

run /global/homes/h/hdumasde/Programs/desi/code/redrock/bin/rrplot
--datatype "desi"
--specfile "/project/projectdirs/desi/datachallenge/reference_runs/18.7/spectro/redux/mini/spectra-64/52/5263/spectra-64-5263.fits"
--rrfile "/project/projectdirs/desi/datachallenge/reference_runs/18.7/spectro/redux/mini/spectra-64/52/5263/redrock-64-5263.h5"