Improve runtime calculations

[michael-berk]

The below functionality could lead to performance issues and should be explored and potentially resolved.

• Each time a query is executed, a new connection is created.
• Benchmarking utils instantiation should only occur once and be shared by threads. It should not be included in the timing code either. If we don't want to deal with thread locks, we can also instantiate once per thread.

[michael-berk]

Based on a quick diff between warehouse query metrics and beaker-reported metrics, beaker overestimates query runtime by 0.2-0.4 seconds. This is significant and should be resolved.

• The overestimate corrupts the elapsed_time results dramatically for short running queries
• The overestimate has relatively little impact on throughput calculations and would be hard to resolve

Best solution: query actual query history instead of timing in-line.
Nest best solution: do the above solutions.

Note that the second bullet should probably be implemented either way.