[clang] [llvm] [RISCV] Allow extra underscores in parseNormalizedArchString and parseArchString. (PR #91532)

[Craig Topper via cfe-commits]

topperc wrote:

> This one seems lazy, avoiding trailing or duplicate underscores is easy.

Yeah I don't really expect anyone to use it. It seemed like an unnecessary complication in our parsing code.

I was considering getting rid of split call before the loop and make the loop work directly on the raw string. Not having to care about extra underscores makes that easier.

https://github.com/llvm/llvm-project/pull/91532