HLSL: Emulate write-to-output on return-from-entry-point, for return value.

This fixes issue #487 and #480.
It also correctly handles output parameters from the entry point.

Test/baseResults/remap.hlsl.sample.basic.everything.frag.out

@@ -13,7 +13,7 @@ Linked fragment stage:
                               Capability Sampled1D
                1:             ExtInstImport  "GLSL.std.450"
                               MemoryModel Logical GLSL450
-                              EntryPoint Fragment 5663  "main"
+                              EntryPoint Fragment 5663  "main" 4045
                               ExecutionMode 5663 OriginUpperLeft
                               Decorate 4727 DescriptorSet 0
                               Decorate 4727 Binding 0
@@ -128,6 +128,8 @@ Linked fragment stage:
              138:   13(float) Constant 1065353216
             1284:   29(fvec4) ConstantComposite 138 138 138 138
              650:             TypePointer Function 13(float)
+            1670:             TypePointer Output 1032(struct)
+            4045:   1670(ptr) Variable Output
             5663:           8 Function None 1282
            24915:             Label
             5830:   1972(ptr) Variable Function
@@ -208,6 +210,7 @@ Linked fragment stage:
                               Store 20158 1284
            19657:    650(ptr) AccessChain 5072 2574
                               Store 19657 138
-           21536:1032(struct) Load 5072
-                              ReturnValue 21536
+           17934:1032(struct) Load 5072
+                              Store 4045 17934
+                              Return
                               FunctionEnd