[llvm] [IR][Float8] Add two kinds float8 IR type (PR #89900)

[Joshua Cranmer via llvm-commits]

https://github.com/jcranmer-intel commented:

Can you add some tests for invalid constants in the IR parser? In particular, I'd like to make sure that the float8e4m3fn is handled properly in the double-hex-constant mode, given that it lacks infinities and has but one NaN value.

https://github.com/llvm/llvm-project/pull/89900