Feature/eps net and summarizer

[xuanfeiren]

In this PR, I added the epsilon net and summarizer features.

1. Add summarizer.py under priority_search.
2. Add a regressor template in module_regressor.py. This could be used to create different types of regressors. In this PR, it is used to get the embeddings of candidates.
3. Add epsNetPS_plus_summarizer.py under priority_search, containing EpsilonNetPS_plus_Summarizer, which is a subclass of PrioritySearch.
4. Modifty the current PrioritySearch. Move "adding the exploration sample rollouts to ModuleCandidate" before proposing new candidates. Now at the propose step, we have information of exploration candidates.