This will allow us to run two different completion kinds and deliver results from both of them.
Also: Compute a unified type context for global lookup. Previously, we always used the expected type context of the last lookup. But really, we should be considering all possible types from all constraint system solutions when computing code completion results from the cache.
Setting the interface type of a variable, just to reset it to a null type is actually really gross. But quite a few methods further down in the generation of code completion results (such as USR generation) need to get a variable’s type and passing them along in a separate map would be really invasive. So this seems like the least bad solution to me.
This eliminates a source of bugs if subclasses of `TypeCheckCompletionCallback` forget to call the superclass’s implementation of `sawSolution` from their overridden method.
