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/hlsl.entry-out.frag

@@ -0,0 +1,12 @@
+struct OutParam {
+    float2 v;
+    int2 i;
+};
+
+float4 PixelShaderFunction(float4 input, out float4 out1, out OutParam out2) : COLOR0
+{
+    out1 = input;
+    out2.v = 2.0;
+    out2.i = 3;
+    return out1;
+}