fiberassign output dependence on environment/compiler

[sybenzvi]

In the survey ops telecon today we reviewed a discovery made in October 2024 by @LNapolitano , @araichoor , and @ashleyjross that fiberassign reproducibility depends sensitively on the environment and the C++ compiler used at NERSC. The issue is documented in two separate threads on slack:

1. A discussion in #f4alss documenting attepts to use DESI environment 23.10, 24.4, and main.
2. Documentation of Anand's custom environment on #data-ops.

The origin of the different results has not been tracked down. Anand's custom environment is a temporary fix; containerizing it would also be a temporary (if prudent) solution.