[clang] [llvm] [RISCV] Add the Sha extension (PR #113820)

[Alex Bradbury via cfe-commits]

asb wrote:

> > I haven't made it so if you enable all extensions that constitute Sha, Sha is implied.
> 
> Great! I think it's most clear if we have "extensions only require things they are made up of" (i.e. all the implications/requirements point to earlier/lesser extensions) and "profiles are how you enforce a lower bound on the set of extensions present" we're going to be in a much more sustainable situation than other architectures where they try to use the presence of a combination of other extensions to lower-bound the set of present extensions.

I've put a note to cover this general topic in the next sync-up call. The issue with not adding it by implication is that any tooling (including LLVM) that queries has to check for all the sub-extensions when it needs to check. One example is that 'B' has retrospectively been added to RVA22. Except right now we don't emit it, and it means that the canonical ISA naming string for an RVA22 processor has arguably changed.

https://github.com/llvm/llvm-project/pull/113820