Concurrent fiber solver

[rgrinberg]

Switch to parallel iteration from sequential iteration. This required switching to a fiber aware cache.

This gives a small but nice boost:

Before:

Benchmark 1: ~/github/ocaml/dune/_build/default/bin/main.exe pkg lock
  Time (mean ± σ):      1.962 s ±  0.069 s    [User: 1.285 s, System: 0.444 s]
  Range (min … max):    1.882 s …  2.086 s    10 runs

After:

Benchmark 1: ~/github/ocaml/dune/_build/default/bin/main.exe pkg lock
  Time (mean ± σ):      1.752 s ±  0.315 s    [User: 1.284 s, System: 0.464 s]
  Range (min … max):    1.535 s …  2.606 s    10 runs

[rgrinberg]

ping @gridbugs or @Leonidas-from-XIV

[gridbugs]

Sorry for the delay reviewing this.