[PATCH] Fixup PHI nodes in LowerSwitch

[Pete]

>>! In D4298#25, @kariddi wrote:
> Reverted the to the getFirstNonPHI() version waiting for more input on the matter

Yeah, from what David said I was wrong about this so good to proceed with the original PHI looping code.

I did take a look in Verifier::visitBasicBlock(BasicBlock &BB) and we don't actually check that PHIs are the first instructions in a BB, or contiguous, but that should be easy to add as David points out its what LangRef says already.

http://reviews.llvm.org/D4298