Track trace with ets tables

[tsloughter]

I think we should investigate the performance of using ETS tables to track traces/spans. This would work similar to how db_connection works now with @fishcakez pool. Setting an heir to be a cleanup process provides the feature of a crashed process can have its open spans closed and finishing can be done with ets:giveaway.

It might be too much overhead on a system with thousands of concurrent traces, but I think worth measuring to be sure.