Parallelize optimization with grid search instead of fmin

[iharsuvorau]

Currently, we are using a Tree-Structured Parzen Estimator (TPE) to guide the search of the best parameters when optimizing the Resource Model and the Control-flow.

Consider the option of using grid search to generate a subset of cases from the search space, and parallelize their computation.

* The TPE cannot be parallelized, as each iteration parameters depends on the loss from the previous ones.