Dynamic send and batch send splits for milestone 1.5

[Gsantomaggio]

1.5 Milestone

The PR introduces a new way to send messages.
We removed the aggregation via timeout and implemented the dynamic send. The goal is to reduce the latency, especially in low-rate use cases.
Low-rate use cases are the most used.
The dynamic send is a wrapper around the BatchSend() that will remain public for the users who want more control during the send.

Test results

PerTest with a limit rate of 50 Messages per second

Given:

 ./perftest --streams stream --publishers 1 --consumers 1 --max-length-bytes 10GB --async-send --rate 50;

With this PR the latency is around 5 ms

Published     40.0 msg/s | Confirmed     40.0 msg/s |  Consumed     40.0 msg/s |  Rate Fx: 50 | Body sz: 8 | latency: 5 ms
Published     41.6 msg/s | Confirmed     41.6 msg/s |  Consumed     41.6 msg/s |  Rate Fx: 50 | Body sz: 8 | latency: 4 ms
Published     42.9 msg/s | Confirmed     42.9 msg/s |  Consumed     42.9 msg/s |  Rate Fx: 50 | Body sz: 8 | latency: 4 ms
Published     43.8 msg/s | Confirmed     43.8 msg/s |  Consumed     43.8 msg/s |  Rate Fx: 50 | Body sz: 8 | latency: 4 ms

Main branch (1.4.x), the latency is around 90 ms ( The timeout for the aggregation )

Published     44.4 msg/s | Confirmed     44.4 msg/s |  Consumed     44.4 msg/s |  Rate Fx: 50 | Body sz: 8 | latency: 108 ms
Published     45.0 msg/s | Confirmed     45.0 msg/s |  Consumed     45.0 msg/s |  Rate Fx: 50 | Body sz: 8 | latency: 106 ms
Published     45.5 msg/s | Confirmed     45.5 msg/s |  Consumed     45.5 msg/s |  Rate Fx: 50 | Body sz: 8 | latency: 103 ms

With the dynamic send, the latency is reduced, especially with a low rate, which is the most common use case.