[llvm] [RISCV] Ensure LPAD alignment for calls to returns_twice functions (PR #177515)

[Kito Cheng via llvm-commits]

kito-cheng wrote:

Yeah, and the awkwardness for current solution (w/o introduction relocation) is the auipc/jalr no longer able to relax into jalr if possible...which is harm to both performance and code size, however return_twice (e.g. setjmp) should be rare, so that might be acceptable solution. 

> I thought there was a neater way to do this, but I think it just moves the hacks around.

I think it's better since we save one full scan for whole function, also including `lpad` within the exact closure seems more semantic correct.

https://github.com/llvm/llvm-project/pull/177515