[llvm] [VectorCombine] Fix duplicate freeze scheduling for shared indexes (PR #213461)

[Aditya prakash jha via llvm-commits]

AdityaOP007 wrote:

@ParkHanbum, I've updated this to use dyn_cast<Instruction>() and only perform the lookup for instruction indices, since constant indices are handled separately in canScalarizeAccess().

https://github.com/llvm/llvm-project/pull/213461