Optimize provenance generation speed

[joinemm]

This PR gives the provenance generator ~60x speed increase when generating recursive provenance for a big system image, from 5 minutes to 5 seconds, without removing any data. In fact, the detail of the provenance is even increased now with the addition of --include-outputs flag, adding previously missing dependencies to the array.

The nix commands themselves are fast but the subprocess.run() calls for every dependency in the loop was causing significant overhead. Upon reading the nix docs I found out that you can pass as many derivations as you want in one nix-store query, and you will get the result for all of them, so we can compute all hashes before the loop in single command.