[Perf] Add option to perf frameworks to specify number of operations instead of time duration

Currently, the perf frameworks only support a time-based duration, like "warmup for 10 seconds, run for 30 seconds".

A common request is to instead specify a number of operations, like "warmup for 100 operations, run for 1000 operations". This is particularly useful when profiling to ensure you collect a reasonable amount of data.

The value doesn't need to be exactly precise. If a user requests say 10k operations, it's fine if we run slightly more or less.

If it's simpler, the value could be per-parallel worker, instead of a global value across all workers. This would eliminate contention on a global variable when every parallel operation is completed. And `parallel = 1` is typically used when profiling.