[playground][server] span latency numbers are affected by number of instances running / number of dataset examples

[Parker-Stafford]

The latencies of spans for each run of a playground (particularly over a dataset) are not representative of the time it took to complete the request because they are affected by the other runs in flight.

Could potentially investigate a way to try to start spans such that they are more representative of the time it took for the actual work of that particular span.