Add Ability to Screen Out Data By Processibility

[WardLT]

As alluded to in #12 , there may be processing conditions which do not produce high-quality films.

We can exclude this data from the model fitting, but could do better by avoiding running these kind of experiments in the first place.

Perhaps we can train a model to predict if a point will be "good or bad," or just exclude points in the search space that are neighbors to the bad point.

[WardLT]

Steps:

• [x] Screen the training data to get the bad points
• [x] Exclude them from the training set
• [ ] Implement the "taboo" logic forbidding similar conditions from being recommended to the robot