BBVectorize: Avoid linear searches within the load-move set

This is another cleanup aimed at eliminating linear searches
in ranges of std::multimap.

No functionality change intended.

llvm-svn: 174858