improve robustness of trace fitting

Night 20231205 arc expid 208156 and 208158 failed due to fiber 885 having a bad trace shift that overlapped fiber 884. This fiber is near a hot column on z1, but most of the fiber is unaffected. The highest wavelengths, however, are masked, resulting in a runaway solution. For now I'll mask this fiber as bad and proceed, but it's unfortunate to permanently discard an entire fiber due to ~100 Angstroms of bad CCD masked data corrupting the trace shifts.

We could have better priors on masked lines to avoid situations like this. In this case, it might even be jumping over to fiber 883 mis-identifying the location of the highest wavelength emission line.


Start from the trace-fitting implementation and the handling of masked CCD data; the issue does not name specific files, tests, or entry points. Reproduce the fiber 885 failure described for exposures 208156 and 208158, then define and verify behavior that avoids a runaway trace solution without discarding an otherwise usable fiber.

Written by the indexing model from the issue text.