[llvm] [RISCV] Handle recurrences in RISCVVLOptimizer (PR #151285)

[Min-Yih Hsu via llvm-commits]

https://github.com/mshockwave commented:

Generally looks good to me, as long as the safe approximation in the dataflow analysis is VLMAX and we never reduce the VL in this process I think we're good. 

> where the height is determined by the number of distinct VL values

Related to this: could we add a test for fault-only-first load?

https://github.com/llvm/llvm-project/pull/151285