Fix bug with mask used for gather/scatter code generation.

We should always use the full mask for this, never the internal mask.
Added tests for this.

tests/scatter-mask-1.ispc

@@ -0,0 +1,15 @@
+
+export uniform int width() { return programCount; }
+
+uniform float a[programCount];
+
+export void f_f(uniform float RET[], uniform float aFOO[]) {
+    int index = aFOO[programIndex]-1; 
+    if (index & 1)
+        a[index] = 1;
+    RET[programIndex] = a[programIndex]; 
+}
+
+export void result(uniform float RET[]) {
+    RET[programIndex] = (programIndex & 1) ? 1 : 0;
+}

tests/scatter-mask-2.ispc

@@ -0,0 +1,19 @@
+
+export uniform int width() { return programCount; }
+
+uniform float a[programCount];
+
+static void foo(int index) {
+    a[index] = 1;
+}
+
+export void f_f(uniform float RET[], uniform float aFOO[]) {
+    int index = aFOO[programIndex]-1; 
+    if (index & 1)
+        foo(index);
+    RET[programIndex] = a[programIndex]; 
+}
+
+export void result(uniform float RET[]) {
+    RET[programIndex] = (programIndex & 1) ? 1 : 0;
+}