Allow using Alias/PTA analyses from LLVM

[mchalupa]

From LLVM 3.9 there are also Andersen's and Steensgaard's analyses in LLVM, so try using them. We'll have more stuff to compare.

[mchalupa]

And also memory dependence analysis

[mchalupa]

For AA, create a wrapper that will take just the partial results (as AA are alias analyses) and propagate as much information as it can into other nodes. For the reset, it just sets unknown pointer. This way we can plug AA into our pointer analysis API.