fix: reliable flash detection at threshold frequencies

[OMARVII]

Summary

• Bug: Maximum (and other) sensitivity modes failed to detect flashes at exact threshold frequencies (~40% miss rate). Reported during manual testing — max mode didn't catch obvious flashes.
• Root cause: The 1-second sliding analysis window was too tight. At boundary-aligned flash cycles (e.g. 3Hz at medium, 1Hz at maximum), the final direction change fell just outside the window.
• Fix: Extended WINDOW_DURATION_MS from 1000ms to 1067ms (one extra frame interval). This guarantees the window captures all transitions at every advertised threshold. No fps increase needed — 15fps is sufficient by Nyquist theorem.
• Added 5 new/updated tests including a realistic 15fps simulation suite that exercises the exact scenario that was broken.

Test plan

• npm test — 70/70 pass (was 65 before, 5 new/updated)
• npm run lint — 0 errors
• Verified 100% detection reliability at all threshold Hz via simulation (1Hz/max, 2Hz/high, 3Hz/medium, 5Hz/low)
• Verified no false positives below threshold (2Hz does NOT trigger medium)

🤖 Generated with Claude Code