[PATCH] An attempt at fixing lifetime extended temporaries in the CFG.

[Jordan Rose]

This mostly seems reasonable to me, except I wonder if we'll want to encode the path to the temporaries by something other than the expression where they were created. I can't remember when they're proper regions and when they're just values (LazyCompoundVals) in the analyzer, but we should be careful to be consistent about the region in which the constructor is run and the region in which the destructor is run.

http://reviews.llvm.org/D4706