Disentangle storage and arithmetic from conversions

Disentangle storage and arithmetic from conversions

Remove the requirement for 16-bit storage from float16 conversions
tests. Add f16 -> f32 and f32 -> f16 tests using only 16-bit storage.

Component: Vulkan

VK-GL-CTS Issue: 1931

Affects:
dEQP-VK.spirv_assembly.instruction.*.fconvert.*float16*
dEQP-VK.spirv_assembly.instruction.*.convertftou.*float16*
dEQP-VK.spirv_assembly.instruction.*.convertftos.*float16*
dEQP-VK.spirv_assembly.instruction.*.convertutof.*float16*
dEQP-VK.spirv_assembly.instruction.*.convertstof.*float16*

Change-Id: I993bdb7a6313cf19ce59b5fa43645f5168ea3cbc