Track zalloc times in microbenchmark

This forms our speed of light for call creation: at some point call
creation is dominated by zeroing memory and all we can do is reduce the
size of the call... let's track where that point is.

test/cpp/microbenchmarks/bm_call_create.cc

@@ -60,6 +60,16 @@ extern "C" {
 
 auto &force_library_initialization = Library::get();
 
+void BM_Zalloc(benchmark::State& state) {
+  // speed of light for call creation is zalloc, so benchmark a few interesting
+  // sizes
+  size_t sz = state.range(0);
+  while (state.KeepRunning()) {
+    gpr_free(gpr_zalloc(sz));
+  }
+}
+BENCHMARK(BM_Zalloc)->Arg(64)->Arg(128)->Arg(256)->Arg(512)->Arg(1024)->Arg(1536)->Arg(2048)->Arg(3072)->Arg(4096)->Arg(5120)->Arg(6144)->Arg(7168);
+
 class BaseChannelFixture {
  public:
   BaseChannelFixture(grpc_channel *channel) : channel_(channel) {}