Consider parallelizing the bounded runs iteration code

[Ratfink]

There are several opportunities to parallelize the bounded runs iteration method.

1. The bounded_runs function essentially performs a depth-first search of the automaton for n time steps. If we instead did a breadth-first search for m < n time steps, then spawned 2^m threads to search the remaining n - m steps, we could take advantage of multicore machines quite effectively.
2. The bounded_runs_iter function runs bounded_runs for different initial automaton locations. We could consider parallelizing this as well, but this would only help after the completion of the "zeroth" iteration (i.e. the call to bounded_runs outside the main for loop). Done!

We should consider implementing one or both of these so the algorithm can take better advantage of compute clusters.