[llvm] [RISCV] Use vsetivli instead of `x0,x0` form to retain SEW/LMUL when AVL is imm (PR #169307)

[Craig Topper via llvm-commits]

topperc wrote:

If the immediate is larger than VLMAX, vsetvli needs to clamp the VL. I'm not sure all implementations would implement this clamping in the decode stage. The vsetvli x0, x0 form can't change VL.

I'm going to consult our architects about how this change will affect our cores.

https://github.com/llvm/llvm-project/pull/169307