[benchmarks] Align dumb_alloc requests to 8 bytes.

And improve the Id histogram.

core/ui.py

@@ -254,11 +254,10 @@ def PrintAst(nodes, opts):
       if 0:
         from _devbuild.gen.id_kind_asdl import Id_str
         from frontend.lexer import ID_HIST
-        total = 0
-        for id_, count in ID_HIST.most_common():
+        for id_, count in ID_HIST.most_common(10):
           print('%8d %s' % (count, Id_str(id_)))
-          total += count
         print()
+        total = sum(ID_HIST.values())
         print('%8d total tokens returned' % total)
 
     else:  # text output

cpp/dumb_alloc.cc

@@ -11,10 +11,11 @@ int gMemPos = 0;
 int gNumNew = 0;
 int gNumDelete = 0;
 
-// https://stackoverflow.com/questions/2022179/c-quick-calculation-of-next-multiple-of-4
+// Align returned pointers to the worst case of 8 bytes (64-bit pointers)
 inline size_t aligned(size_t n) {
-  //return (n + 7) & ~7;
-  return (n + 15) & ~15;
+  // https://stackoverflow.com/questions/2022179/c-quick-calculation-of-next-multiple-of-4
+  return (n + 7) & ~7;
+  //return (n + 15) & ~15;
 }
 
 // This global interface is silly ...