Improve per-iteration leak scan fuzzer logging for readability

This is an improvement for the per-iteration leak scan fuzzer feature.
Reference: #1797

The current error logging implementation is somewhat confusing, as it reports all previous leaks in every subsequent iteration. For example, a single memory leak found in Iteration 1 is reported in Iteration 2, Iteration 3, and so on..

The desired behavior is that the report for any given iteration N should only include new leaks discovered in iteration N.

I think the best approach to solve this is to take the difference between the updated error counters and the saved ones at the checkpoint.