Make bucketing standalone, for reuse by other tools

[Chris Warburton]

Rather than having separate theory exploration tools (HashSpec, MLSpec,
etc.) we can instead make bucketing into a generic pre-processor, which
can be tested on different tools.

In particular, timing how long the bucketing takes isn't particularly
interesting: we can assume it's pretty darn fast. If not, we can just
benchmark it separately, rather than complicating the timing of the
exploration phase.