Fix race condition in start-dumb-udev.sh causing repeated Xorg restarts

[qdii]

start-dumb-udev.sh restarts Xorg once when it notices a Sunshine virtual input device (or an evdev passthrough device) appear, so Xorg can pick it up. It remembers having done this with a flag file, and only allows itself to restart Xorg again once that device has gone away.

The problem is a race between two things happening on their own timelines: the watcher checks for the device once a second, while Sunshine removes and recreates its virtual devices independently (for example while a client keeps retrying a connection). If the watcher's check happens to land in the brief moment between Sunshine removing the old device and creating the new one, it thinks the device is gone for good and resets its flag. The moment the new device appears a second later, it looks like a brand new device to the watcher, which restarts Xorg again. A flaky client connection can turn this into a continuous Xorg restart loop, taking down the whole desktop session.

The fix: don't trust a single absent check. The watcher now needs to see the device missing for several consecutive seconds before it treats it as actually gone (configurable via ABSENCE_DEBOUNCE_SECONDS, default 10). A quick remove-then-recreate no longer falls inside that window, so it's no longer mistaken for a fresh device needing a fresh restart.