Fold the set_events call into evaluate_pdf

[mjlarson]

It might be more natural (and easier to work with) if the KingSpatialLikelihood.evaluate_pdf were to be able to call set_events itself. To do this, we'd need to

• Set up a cache of source RA/dec
• For each call to evaluate_pdf, take the events + source locations
• On each call to set_events, check if the events have changed
• If the events HAVE changed: empty the source RA/dec cache, empty the event pvalue cache, add the current source coordinates to the cache, refill the event pvalue cache, and store the current source index
• If the events HAVE NOT changed and the source RA/dec combination EXISTS in the cache: update the source index
• If the events HAVE NOT changed and the source RA/dec combination DOES NOT EXIST in the cache: calculate the event pvalue cache and update the source index.

There might have to be some way to ensure that we don't end up with crazy storage use. Is this a case where we should adopt scipy.sparse.sparse_array?