Benchmarking: Adjust the load profile to be flat across time

Currently, the k6 tests do not maintain a flat load profile over time. This makes the results a little hard to understand. Let's adjust the test so that it's a flat profile in the volume of rpc calls over time.

Also, it was noted by @msmiley that we don't want to push the box to CPU saturation since we are moving the bottleneck.

Instead, let's separate out two separate benchmarks--a generic workload that remains flat at a sustained high load but not pushing the box to CPU saturation.

Then, we should fashion a separate test that aims to trigger different bottlenecks we have discovered.