Speed up getGroundTruths

[Chris Warburton]

It's really slow, taking several seconds per run. That's unacceptable if
we want to get the ground truths before and after bucketing for
thousands of samples.

It's currently written in Racket, since the required data and functions
are provided by tebenchmark. The obvious thing to do would be:

 1) Port it to Haskell, since it has lower overhead and faster startup.
 2) Switch to fast-lookup datastructures, even if they're slow to build
 3) Move file reading and datastructure building to compile time with TH