Improve async appender performance

[swebb2066]

This PR changes AsyncAppender implementation to use a preallocated ring buffer and is inspired by LMAX Disruptor

The benchmark comparison to master ( benchmark/tools/compare.py --no-color benchmarks /tmp/old.json /tmp/new.json ) shows the following improvement on Ubuntu (gcc 9):

Benchmark	Time	CPU	Time Old	Time New	CPU Old	CPU New
Appending int+float using MessageBuffer, pattern: %m%n	-0.0288	-0.0289	966	938	966	938
Appending int+float using MessageBuffer, pattern: %m%n/threads:4	+0.0164	+0.0157	425	432	1698	1725
Appending int+float using FMT, pattern: %m%n	-0.0621	-0.0621	543	509	543	509
Appending int+float using FMT, pattern: %m%n/threads:4	+0.0034	+0.0039	380	381	1511	1517
Async, Sending int+float using MessageBuffer	-0.316	-0.3159	2171	1485	2171	1486
Async, Sending int+float using MessageBuffer/threads:4	-0.527	-0.4956	847	400	3175	1601