[llvm] [X86] Fold vpextrq $1 from a spilled vector into a direct memory load (PR #203339)

[Simon Pilgrim via llvm-commits]

RKSimon wrote:

I still think we're going about this the wrong way - we should be spilling the scalar not the whole vector. We're kludging it because everyone is afraid of regalloc and spiller :/

https://github.com/llvm/llvm-project/pull/203339