[llvm] RISCVAsmParser: Don't treat operands with relocation specifier as parse-time constants (PR #133377)

[Fangrui Song via llvm-commits]

MaskRay wrote:

> Happy to defer to Sam's review of the patches, but I wanted to give an explicit LGTM on the direction and to say thank you for taking the time to look at how we clean up our MC code. Writing this evaluation logic was a useful exercise at the time, but I agree it's unnecessary complexity at this point.

Thanks for the LGTM and the kind words! I have recently spent quite some time on AsmParser across various targets and realized what we could simplify... 
(BTW, thanks for making a great choice by embracing `MCTargetExpr` early on instead of `MCSymbolRef::VariantKind`-back in 2017, I believe only AArch64, PowerPC's @l/@ha (with significant tech debt), and MIPS (https://reviews.llvm.org/D19716; however, also with significant tech debt) were using it. That choice helped prevent a lot of complexity.)

https://github.com/llvm/llvm-project/pull/133377