Reviewer roulette should give stable recommentations
Currently, reviewer roulette runs every time someone pushes to their merge request. Each time it runs, it makes a random selection of reviewer and maintainer. This means the selections are not stable - they change every time a push happens.
In the introducing MR, we discussed this a little, and I wasn't sure whether it would be annoying or not, so we left it at the simplest iteration.
Having experienced it for a couple of days, it is starting to annoy me, so perhaps it should change?
Implementation-wise, we spoke of seeding the random number generator to get a stable recommendation, but that has a number of limitations, and doesn't guarantee stability in some circumstances.
I wonder if we could read the recommendations made in the last run, if we mark up the table we're outputting to the GitLab comment appropriately. We can then re-use any suggestions that already exist, while adding new suggestions if the list of categories change and some are unfilfilled.