Add some comments clarifying that the GEP analysis for vector GEPs is

deeply suspicious and likely to go away eventually. Also fix a bogus
comment about one of the checks in the vector GEP analysis. Based on
review from Duncan.

llvm-svn: 163894