PowerPC mulhdu lifting incorrect due to core analysis bug #8682

Originally posted by @zznop in #8665 (comment)

Not advocating that you should fix this as it's not your issue, but this code does expose it: for mulhdu with both operands equal to 1 << 32 the expected high half is 1. But instead our core analysis folds the 128-bit product to 0 leaving seemingly incorrect HLIL.