float16_t does not take GLSL precisions and SPIR-V does not support RelaxedPrecision on float16_t. Fixes #2894