[chttp2] Dont send ping-acks with large data frames (#34583)

If we send a large amount of data along with an ack, then we bundle that
ack in with a crypto frame that might be very large. This causes two
problems:
1. we need to unencrypt a large frame (up to 1MB with ALTS) before we
can ack the ping, and on a slow connection this could take some time.
2. we need to do all the crypto work up front to send that ping ack,
also creating lots of work.

I think there's an equivalent fix needed on the ping send side, but less
urgent because it's not currently causing flakiness!

src/core/ext/transport/chttp2/transport/writing.cc

@@ -227,11 +227,6 @@ static void report_stall(grpc_chttp2_transport* t, grpc_chttp2_stream* s,
   }
 }
 
-// How many bytes would we like to put on the wire during a single syscall
-static uint32_t target_write_size(grpc_chttp2_transport* /*t*/) {
-  return 1024 * 1024;
-}
-
 namespace {
 
 class CountDefaultMetadataEncoder {
@@ -302,6 +297,10 @@ class WriteContext {
   }
 
   void FlushPingAcks() {
+    if (t_->ping_ack_count == 0) return;
+    // Limit the size of writes if we include ping acks - to avoid the ack being
+    // delayed by crypto operations.
+    target_write_size_ = 0;
     for (size_t i = 0; i < t_->ping_ack_count; i++) {
       grpc_slice_buffer_add(t_->outbuf.c_slice_buffer(),
                             grpc_chttp2_ping_create(true, t_->ping_acks[i]));
@@ -328,7 +327,7 @@ class WriteContext {
   }
 
   grpc_chttp2_stream* NextStream() {
-    if (t_->outbuf.c_slice_buffer()->length > target_write_size(t_)) {
+    if (t_->outbuf.c_slice_buffer()->length > target_write_size_) {
       result_.partial = true;
       return nullptr;
     }
@@ -357,6 +356,7 @@ class WriteContext {
 
  private:
   grpc_chttp2_transport* const t_;
+  size_t target_write_size_ = 1024 * 1024;
 
   // stats histogram counters: we increment these throughout this function,
   // and at the end publish to the central stats histograms