break out different component functions in Gibbs sampler

[TommyJones]

modularize code. make it less unwieldy. Early performance issues were due to my compiler, not function calls.

[TommyJones]

Components include:

1. Initial data prep
2. Recalculate probabilities
3. Sample topics (which would decrease/increase counts and call recalculate probabilities)
4. Calculate likelihood
5. Optimize alpha
6. Aggregate counts after burnin (sum Cd, Ck, Cv across iterations)
7. Cleanup results for return

The architecture would call each of these functions from a main function. See #11 for a description of how it works once parallelized

[TommyJones]

Actually, now I think that initial data prep and cleanup results for return could just be part of the main function.

[TommyJones]

Ended up with recalculate probabilities and sample topics in one function.

Done with 9119b3e