[llvm] [RISCV] Combine vslidedown_vl with known VL to a smaller LMUL (PR #66267)

[Luke Lau via llvm-commits]

lukel97 wrote:

> This patch isn't sound. Just because the VL is known doesn't mean the slidedown offset is <= VL. The elements that will be written to the lower elements of the destination may come from the high LMUL part of the input.

Sorry you're right, I think this should be `getSmallestVTForIndex(OrigVT, VL + Offset, DL, DAG, Subtarget)`

https://github.com/llvm/llvm-project/pull/66267