Log query errors through tracing crate

[pkolaczk]

No description provided.

[pkolaczk]

CC: @vponomaryov this is unfinished, but this should allow you to intercept query errors while running. Let me know what you think.

[vponomaryov]

CC: @vponomaryov this is unfinished, but this should allow you to intercept query errors while running. Let me know what you think.

Lots of stress tools print out errors into the same output as the general stats.
Cassandra-stress is one of such popular tools.
We find this approach useful and prefer it.
Having a separate file for stats and another one for errors just makes us do data aggregation or spend more time for matching time slots when something (controlled disruptions) goes wrong (always).
Errors are valuable in conjunction to the growing latencies of OK stats too.

So, lots of new features are interesting to have.
But, from our point of view, we get too big regressions which make us think about switch from sync fork to origin to backport interesting changes to our fork.