[C++] 'dumb' search functions over hyperparameters should return found_flag like all other optimizers

[suntzu86]

EvaluateLogLikelihoodAtPointList and LatinHypercubeSearchHyperparameterOptimization in gpp_model_selection_and_hyperparameter_optimization.hpp do not pass back a "found_flag" term, unlike every other optimizer/evaluation function in optimal_learning.

For consistency, we should change this.

1. EvaluateLogLikelihoodAtPointList should have a found_flag output that is read from io_container right before return.
2. LatinHypercubeSearchHyperparameterOptimization should have a found_flag output that is passed straight to EvaluateLogLikelihoodAtPointList
3. InitializeBestKnownPoint should have check_all_points dropped and it should ONLY be called from places that don't invoke the NullOptimizer.
4. EvaluateLogLikelihoodAtPointList should stop calling InitializeBestKnownPoint. Instead, build io_container with the first multistart point (arbitrary) and set its value to -std::numeric_limits<double>::max() (or -infinity()?)