[mycpp/runtime] Make Pools about 16 KiB each

Rather than 4 KiB each.  The 16 bytes of "slack" actually seems useful
for the malloc() header.

Showed up as a VERY slight improvement on benchmarks2/uftrace!

Should also look at memory usage, though in theory it can't change much.

cpp/obj_layout_test.cc

@@ -143,9 +143,18 @@ TEST malloc_address_test() {
   // 4. diff = 8
   // 5. diff = 8
 
+  // 2023-08: If I pass 4096, I get 4112, so 16 byte diff
+  // 2023-08: If I pass 4080, I get 4096
+
+  // int alloc_size = 24 * 682;  // 16368 is close to 16384 - 16 bytes again
+  int alloc_size = 48 * 341;  // heap 2 is the same size
+
+  //int alloc_size = 4080;
+  //int alloc_size = 1;
+
   char *p[20];
   for (int i = 0; i < 20; ++i) {
-    p[i] = static_cast<char *>(malloc(1));
+    p[i] = static_cast<char *>(malloc(alloc_size));
     if (i != 0) {
       char *prev = p[i - 1];
       log("%2d. diff = %d", i, p[i] - prev);

mycpp/mark_sweep_heap.h

@@ -264,10 +264,12 @@ class MarkSweepHeap {
   double total_gc_millis_ = 0.0;
 
 #ifndef NO_POOL_ALLOC
-  // 4096 / 24 bytes = 170 cells (rounded), 4080 bytes
-  // 4096 / 48 bytes =  85 cells (rounded), 4080 bytes
-  Pool<170, 24> pool1_;
-  Pool<85, 48> pool2_;
+  // 16,384 / 24 bytes = 682 cells (rounded), 16,368 bytes
+  // 16,384 / 48 bytes = 341 cells (rounded), 16,368 bytes
+  // Conveniently, the glibc malloc header is 16 bytes, giving exactly 16 Ki
+  // differences
+  Pool<682, 24> pool1_;
+  Pool<341, 48> pool2_;
 #endif
 
   std::vector<RawObject**> roots_;