[llvm] [APFloat] Fix sign bit corrupting exponent in unsigned FP8 formats (PR #223286)

[Hung-Kuan Tseng via llvm-commits]

Tim096 wrote:

Thanks for the cc, @tgymnich. I can take the follow-up: the asserts when converting a negative Inf or NaN to `Float8E5M3FNU`, and `subtract` giving a negative result in an unsigned format. I'll send it once this one has landed.


https://github.com/llvm/llvm-project/pull/223286