trace: add flush worker and optimize memory usage

[sreimers]

This reduces trace memory usage from ~137 MByte (1.000.000 entries) to 13,7 MByte (100.000 entries). Every second the trace buffer is flushed asynchronously if more than 1000 entries reached (TRACE_FLUSH_THRESHOLD).

All values can be overriden on compile time:

-DCMAKE_C_FLAGS="-DTRACE_BUFFER_SIZE=10000 -DTRACE_FLUSH_THRESHOLD=100 -DTRACE_FLUSH_TMR=1000"

TRACE_FLUSH_TMR is in ms.

[cspiel1]

We should be careful that the flush does not cause massive CPU load. This may influence the result.

[sreimers]

Flush should mostly I/O bound and runs on a different thread, but I can think about some optimizations, like running tmr more periodically and flush smaller amount of entries.