Don't do updatePrecision on float16_t operations

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

Fixes #2894

Test/spv.float16NoRelaxed.vert

@@ -0,0 +1,16 @@
+#version 450
+#extension GL_KHR_shader_subgroup_vote: enable
+#extension GL_EXT_shader_subgroup_extended_types_float16 : enable
+layout(set = 0, binding = 0, std430) buffer Buffer1
+{
+  uint result[];
+};
+
+void main (void)
+{
+  uint tempRes;
+  float16_t valueNoEqual = float16_t(gl_SubgroupInvocationID);
+  tempRes = subgroupAllEqual(valueNoEqual) ? 0x0 : 0x10;
+  result[gl_VertexIndex] = tempRes;
+}
+