Add benchmarks for nix-ninja generating derivations to compile NixOS/Nix

[elpdt852]

The code around doing the topological scheduling of tasks already uses threads so that each target does dependency inference concurrently. However, there's still a lot of low hanging fruit. Let's add benchmarks for the derivation generating portion so we can speed this up empirically.

Currently Nix 2.27.1 takes ~2m12s to generate all the derivations to compile src/nix/nix:

real	2m12.433s
user	7m14.126s
sys	2m55.770s