Messages are serialized to multiple buffers, intsead of yield each
buffer one by one, create single buffers that contain the whole
serialized message. This greatly improves transport performance as
writing one big buffer is a lot faster than writing lots of small
buffers to network sockets etc.

Before:

```
testing 0.40.x-mplex
sender 3276811 messages 17 invocations
sender 6553636 bufs 17 b 24197 ms
105 MB in 32 B chunks in 24170ms
```

After:

```
testing 0.40.x-mplex
sender 3276811 messages 1638408 invocations
1638411 bufs 68 b 8626 ms
105 MB in 32 B chunks in 8611ms
```