Properly cache assertions in static learning preprocessing pass. (#7242)

Properly cache assertions in static learning preprocessing pass. (#7242)

The cache is user-context dependent and guarantees that ppStaticLearn is only called once per assertion in a given context - if assertions are popped from the context and added again ppStaticLearn is called again on that assertion. Further, the preprocessing pass now also handles the splitting of top-level AND terms. Before that each theory was responsible for splitting and caching. Refactoring the corresponding ppStaticLearn theory methods will be part of future PRs.

Marking do-not-merge until cluster runs finished.