[mlir] LLVM: Convert f8 types to i8 when converting to LLVM types

[ftynse]

Is this the right thing to do? What happens if we have arith.addf %0, %1 : f8, does it get converted to llvm.fadd i8 that will assert later on translation? In any case this needs a test.

mlir/lib/Conversion/LLVMCommon/TypeConverter.cpp

 Type LLVMTypeConverter::convertIndexType(IndexType type) {
   return getIndexType();
 }
 
 Type LLVMTypeConverter::convertIntegerType(IntegerType type) {
   return IntegerType::get(&getContext(), type.getWidth());
 }
 
-Type LLVMTypeConverter::convertFloatType(FloatType type) { return type; }
+Type LLVMTypeConverter::convertFloatType(FloatType type) {
+  // TODO: Do not convert `f8` types to `i8` once LLVM will fully support them.
+  if (type.isFloat8E4M3FN() || type.isFloat8E5M2())
+    return IntegerType::get(type.getContext(), 8);
+  return type;
+}
 
 // Convert a `ComplexType` to an LLVM type. The result is a complex number
 // struct with entries for the
 //   1. real part and for the
 //   2. imaginary part.
 Type LLVMTypeConverter::convertComplexType(ComplexType type) {
   auto elementType = convertType(type.getElementType());
   return LLVM::LLVMStructType::getLiteral(&getContext(),