bugfix: stdstars and relsymlink

[sbailey]

This PR fixes #1890, i.e. a bug introduced with relative symlinks in PR #1888 that broke the symlinks creation for multi-exposure stdstar fitting. This PR adds two redundant fixes:

1. relsymlink(src, dst) now accepts relative paths for src instead of requiring two absolute paths and deriving the relative path. If src is a relative path, it is relative to dirname(dst), not relative to the current directory (same behavior as os.path.symlink). I added tests to confirm this behavior.
2. desi_proc_joint_fit no longer derives the relative path, leaving it to relsymlink instead. In this case desi_proc_joint_fit was deriving the correct path anyway so fix (1) would have been sufficient, but it has the possible future fragility of /dvs_ro/cfs vs. /global/cfs, so I also updated it to let relsymlink standardize how relative paths are computed, since os.path.relpath isn't sufficient for all cases due to read-only vs. read-write mounts of the same directories.

Extra: I also updated some log.debug(...) messages to use log.debug("format string %s", blat) instead of log.debug(f"format string {blat}") so that it doesn't spend time evaluating the string if it isn't going to log it. That has mattered in the past where debug statements were in loops and ended up taking a significant fraction of the compute time. In this case it probably doesn't matter, but since I was updating a log.debug message anyway for the link variable name, I also updated the formatting method and then did the same for several others nearby.

Test run in /global/cfs/cdirs/desi/users/sjbailey/spectro/redux/p8r night 20201214, in particular tile 80605 exposures 00067710, 00067711, 00067712, 00067713.

[akremin]

The changes look good thanks. There was a failure in cholesky_solve, unrelated to this PR. I am rerunning the test now then I can merge.