Do not store results of self-profile benchmark runs into the DB

[Kobzol]

Currently, if you request --self-profile gathering in the collector, it will gather self profiles for the first iteration of a single benchmark, and then ignore the flag for any further iterations. It now stores gathered metrics (e.g. instruction count) into the DB for the first iteration, which is not a good idea - these values are 20-30% higher than the baseline (because self-profile is enabled), and they needlessly occupy space in the DB, and can cause weird results like this.

We should just ignore the gathered metrics for the first iteration of the benchmark run if self profile gathering is enabled.