We should be more aggresive when deleting old traces

[kikito]

Currently we delete traces based on the number of traces alone.

This might not be enough: a bunch of big traces can consume all the available memory and make the monitor fail - this has caused us trouble with 1 user already. We should be erasing by memory consumption instead of by number. And we could probably remove really old traces (i.e. 1 week old).

[mikz]

Deleting them by time still does not solve the issue. Only way how to do this properly is to count the memory usage. We already have internal ticket about this.