[llvm] [RISCV] Fix musttail with indirect arguments by forwarding incoming pointers (PR #185094)

[Xavier Roche via llvm-commits]

xroche wrote:

> I think if you would feel more confident testing this in Qemu, then you should do what you need to feel confident :)

I crafted (still with the help of our Claude friend) locally a qemu set of tests:
- forwarded indirect arg (the original PR fix)
- computed indirect arg (the case you caught
- swapped indirect args (the `OrigArgIndex` asymmetry in `OutputArg` vs `InputArg`
- musttail behind a branch (the SDValue lifetime regression you flagged)
- 10-arg stack spill (your earlier ask)
- sret + musttail
- indirect + spill on the same call

I'd be happy to push it on a public repo if this can be of any value to anyone!

https://github.com/llvm/llvm-project/pull/185094