Should explicit conversion with half exist?

[rjodinchr]

Documentation says that no explicit conversion is provided for half:

• https://registry.khronos.org/OpenCL/specs/3.0-unified/html/OpenCL_C.html#explicit-conversions
• https://github.com/KhronosGroup/OpenCL-Docs/blob/main/man/static/convert_T.txt

But the header in clang is defining it in both direction (input and output) for every type:

• https://github.com/llvm/llvm-project/blob/main/clang/lib/Headers/opencl-c.h

I see half conversion being used in tensorflow lite:

• https://github.com/tensorflow/tensorflow/blob/6dda17ce00d3042e1e2a6c2be252e06bbc3b3846/tensorflow/lite/delegates/gpu/cl/cl_operation.cc#L86

[bashbaug]

Recording discussion from working group teleconference: this is an issue merging with the cl_khr_fp16 spec. Need to document that this functionality is required, but only when the extension is supported.