LIFO ordering for queued sync_info messages

[acco]

Problem

1. Populate a producer's buffer
2. Queue multiple sync_info/async_info messages
3. Drain the buffer
4. sync_info messages are delivered - but they are delivered LIFO instead of FIFO

Example If I add the following test, it fails:

 test "delivers multiple info to producer in FIFO order" do
    {:ok, producer} = Counter.start_link({:producer, self(), buffer_size: 3})
    {:ok, consumer} = Forwarder.start_link({:consumer, self()})

    # Fill buffer so info messages get queued
    Counter.sync_queue(producer, [:a, :b, :c])
    GenStage.sync_info(producer, :first)
    GenStage.sync_info(producer, :second)
    GenStage.sync_info(producer, :third)
    refute_received :first
    refute_received :second
    refute_received :third

    # Subscribe and consume events
    GenStage.sync_subscribe(consumer, to: producer)
    assert_receive {:consumed, [:a, :b, :c]}
    # Skip subscription messages
    assert_receive {:consumer_subscribed, _}
    assert_receive {:producer_subscribed, _}

    messages =
      for n <- 1..3 do
        receive do
          msg -> msg
        after
          100 -> flunk("Did not receive expected number of messages (got #{n - 1})")
        end
      end

    assert messages == [:first, :second, :third]
  end

  1) test info delivers multiple info to producer in FIFO order (GenStageTest)
     test/gen_stage_test.exs:1064
     Assertion with == failed
     code:  assert messages == [:first, :second, :third]
     left:  [:third, :second, :first]
     right: [:first, :second, :third]
     stacktrace:
       test/gen_stage_test.exs:1093: (test)

I assume FIFO here is intended behavior? If so, happy to find root cause and fix.

This was Claude's first suspicion, I can look with fresh eyes:

[josevalim]

Yes, correct, info messages should always be in FIFO. But I am not sure if reverse is correct. We may need to queue some events after the sync_info as well and guarantee they are correct too.