[clang] [llvm] [RISC-V][RVY] Fix round-trip and order of arch strings with 'y' (PR #227865)

[Jessica Clarke via llvm-commits]

jrtc27 wrote:

No, because that’s not a valid arch string. rv64liya0p99 would be ok, but you cannot list two base ISAs in the same arch string.

There is an open question about what the canonical expanded arch string is in a long base ISA name world. I was thinking that through last night, and am not sure what the right thing is here. I feel it might be to do the opposite of what we do for versions, which would be to remove all default parts of the base name, just so the same name remains canonical even in the presence of new components being supported. Otherwise you force all other tools to know about that base ISA component the moment you start emitting it, even if it’s the default one that was previously assumed.

https://github.com/llvm/llvm-project/pull/227865