Investigate runtime estimator panic after PR#3505

[ailisp]

No description provided.

[ailisp]

The panic is due to randomness selecting account when add some key value to storage, and in next step metric doesn't have required keys in storage because randomly select other accounts which did not have add key value step. Therefore some metric (trigger when delete key value in storage) didn't trigger at all, and cause either stat calc overflow, or some key doesn't present in metric stat at all and panicked (not every time panicked, and may panicked for one of these two places). @evgenykuzyakov observed it didn't happen when there's a warmup iter, but i think that's warmup iter increase total number of txns and has a higher chance that one time account has key value to delete and thus the delete from store metrics is recorded. And obviously this is not correct as most of time delete from store metrics is not triggered.

So back to PR, #3505 itself has no issue, above described is a long existing issue in runtime params estimator and described in issue #3601