TraceStack: when visiting a range preallocate the elements.

Improves performance of raytrace.py from about 10secs to 7secs.

src/gc/collector.cpp

@@ -55,6 +55,7 @@ void TraceStackGCVisitor::visit(void* p) {
 }
 
 void TraceStackGCVisitor::visitRange(void** start, void** end) {
+    stack->reserve(end-start);
     while (start < end) {
         _visit(*start);
         start++;

src/gc/collector.h

@@ -59,6 +59,10 @@ class TraceStack {
             return v.size();
         }
 
+        void reserve(int num) {
+            v.reserve(num + size());
+        }
+
         void* pop() {
             if (v.size()) {
                 void* r = v.back();