[llvm] [RISCV] Recurse on second operand of two operand shuffles (PR #79197)

[Luke Lau via llvm-commits]

lukel97 wrote:

Looks like this missed the llvm 18 cutoff, do we want to backport this once it lands so that we're not leaving #79180 in on its own?

https://github.com/llvm/llvm-project/pull/79197