[RFC][clang] - support new clang builtin __arithmetic_fence to control floating point optimization, and new clang option -fprotect-parens

I’ve posted a clang RFC here https://reviews.llvm.org/D100118 proposing a new clang builtin to control floating pont optimization at the expression level. Looking for your feedback, thank you.

It’s dependent on the llvm patch here https://reviews.llvm.org/D99675

The patch isn’t complete so not yet ready for code review. The bulk of the work is to llvm, and the clang patch is straightforward.