Measure tracing overhead

[tsloughter]

After #49 is merged it would be good to be able to include some information on the overhead of operations and at what point (rate of spans being created/modified) does ets lock contention start to degrade performance.

This will also provide a starting point for investigating support for very heavy loads in the future -- such as sharding ets tables or using a pool of processes for ets manipulation instead of doing it inline of the traced process.