release(4.4.1): per-request CPU fix for the scan budget and the performance monitor #174

I am releasing 4.4.1 with the per-request CPU fix from #172 / #173: the scan time budget charges CPU once per pass instead of once per pattern, the event-loop yield slice spans the whole request, and the performance monitor statistics are cheaper. No configuration or detection result changes. Core engines stay in lockstep, so every family ships 4.4.1.