[llvm] [AArch64] Use vecshiftL64 instead of vecshiftR64 to match scalar SLI imm. (PR #139904)

[Paul Walker via llvm-commits]

paulwalker-arm wrote:

Sorry @rj-jesus, it seems this is a more recent regression caused by https://github.com/llvm/llvm-project/commit/e762baffd9f115106f861a236c56093c85dfbc93 which land without review :(

https://github.com/llvm/llvm-project/pull/139904