[x86] Delete some extraneous logic from the new vector shuffle lowering.

Nothing was relying on this and there are potentially some edge cases
that it would not be correct under. Removing it seems better than trying
to "fix" it as nothing was relying on it.

llvm-svn: 218755