Switch back to a stable sort

[lqd]

I was testing rayon's parallel sort on Relations when I noticed we were using std's sort_unstable even though we (albeit not completely 100% sure why) saw the stable sort perform better. So let's bring it back until we have more measurements/benchmarks :)

This locally improves, on clap:

• the DatafrogOpt variant from 5.8s to 5.4s
• the Leapfrog variant from 4.2s to 4s
• the LocationInsensitive variant from 800ms to 410ms (this one convinced me to open the PR as a combination of it with leapfrog is likely going to be the way to go)

[frankmcsherry]

There are places where the stable sort is def better, because it is a merge sort that looks for sorted runs and if they exist it is better than the pattern defeating quicksort (the unstable implementation). The stable sort has run-time guarantees, which is another mark it its favor. Seems good to me. :)

[lqd]

The hype update :) — numbers on unreleased software:

• the LocationInsensitive leapfrog variant goes from 139ms to 127ms