spec flat arc association tweak

[phirstgemini]

When they do science-flat-arc-science-flat-arc-science... all in the same obsid, we should have the calibration rule constrain to that obsid within the same night and we should maybe be matching on closest gravity vector rather than closest in time. It's sub-optimal when they do that but the first or last science frames get matched with flarcs from other observations just because they're marginally closer in time.

I think with GMOS they usually do flarc-science-flarc so it's not usually an issue, but with GNIRS not so much.

Reported example:
https://archive.gemini.edu/searchform/GN-2016B-Q-57-59