SPIR-V validation error when using dFdx/dFdy on float16_t type

Sample code:
```glsl
#version 450
#extension GL_EXT_shader_explicit_arithmetic_types_float16 : require

layout(location = 0) in float input_var;
layout(location = 0) out float output_var;

void main()
{
float16_t input_var_float16 = float16_t(input_var);
float16_t input_var_ddx = dFdx(input_var_float16);
output_var = float(input_var_ddx);
}
```

Validation error printed through `--spirv-val`:
```
error: SPIRV-Tools Validation Errors
error: Result type component width must be 32 bits
%16 = OpDPdx %half %15
```

This operation doesn't seem to be supported by the SPIR-V spec.

Reproduce the sample GLSL program and inspect the SPIR-V output produced when running --spirv-val. Check the SPIR-V specification and validation behavior for dFdx/dFdy with float16_t; done means the unsupported or incorrect behavior has a documented, agreed resolution and the validation result is consistent with it.

Written by the indexing model from the issue text.