fix: make sample size reporting in metrics more congruent/deterministic

[josephcopenhaver]

with the enqueue intent/reality.

When adjusting the interval task size over time there has been a known indeterminate behavior. If the user defined configuration manager of the runner is interested in adjusting task count per interval from 3, to 9, to 27 the metrics reports may go from reporting groups of 3 to groups of 27 and skip over one or more groups of the middle task count intent size (9).

This change adjusts the logic to be 1-1 or 100% deterministic with load generation task count blocks without sacrificing much efficiency on reporting statistics per interval given existing back-pressure logic.

This is achieved by using a single slice of ints in a circular buffer fashion that takes constant time to read and write. Overflows of this buffer are not possible given other assurances that exist in the code.