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

[Joshua Cranmer via llvm-commits]

================
@@ -3847,6 +3847,14 @@ Floating-Point Types
    * - Type
      - Description
 
+   * - ``float8e5m2``
+     - 16-bit floating-point value(2-bit significand)
+
+   * - ``float8e4m3fn``
+     - 16-bit floating-point value(3-bit significand), there are no infinity
+       values, and NaN is represented with the exponent and mantissa bits set
+       to all 1s
----------------
jcranmer-intel wrote:

Are there any references to the definition of the type you can provide, like `bfloat` provides references to hardware ISAs that contain more detail?

(Presumably the reason to finally get around to adding these types in LLVM IR is to enable hardware instructions, so references to hardware ISAs are ideal).

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