Rank fallback reviewers by the count of files that the fallbacks apply to

[imsky]

if there are many path fallbacks that apply to a PR where 70% of the files have a single applicable fallback, contrary to user expectations, the likelihood that a reviewer from the single fallback is chosen is 1/(all reviewers from all fallbacks). one improvement could be to sort fallbacks by applicable file count and drop the lowest "scoring" ones