[llvm] [RISCV] Fixed redundant add instructions generated during the fold of vscale add in rvv. (PR #225266)

[via llvm-commits]

FoolgryGamer wrote:

> Please use `llvm/utils/update_llc_test_checks.py` to update all these failed tests. @FoolgryGamer

I updated the 33 failing tests; among them, `CodeGen/AArch64/sve-calling-convention-byref.ll` was not automatically generated by `llvm/utils/update_llc_test_checks.py`, so I updated the remaining 32 files.

This PR caused a performance regression in the test cases for the AArch64 architecture. All of them resulted in an increase in the instruction count.

I will attempt to implement this optimization solely for the RISCV.

https://github.com/llvm/llvm-project/pull/225266