[llvm] [AArch64] Fold addv(ctpop) to ctpop if the operand is known to be 8-bit (PR #159086)

[Hongyu Chen via llvm-commits]

XChy wrote:

Hmm, maybe we can generalize the fold to `addv(a) -> a`.

https://github.com/llvm/llvm-project/pull/159086