Fix connection throttle cleanup wiping the whole tracker

[gunjanjaswal]

Problem

The connection throttle cleanup in ServerHandshakePacketListenerImpl wipes the entire throttleTracker every ~200 connections. The cleanup predicate compares a stored timestamp against the throttle duration:

throttleTracker.values().removeIf(time -> time > connectionThrottle);

time is an epoch-millisecond timestamp (~1.7e12); connectionThrottle is a duration (default 4000ms). So time > connectionThrottle is always true and every entry is removed. The check a few lines above already does it correctly:

currentTime - throttleTracker.get(address) < connectionThrottle

Within a cleanup cycle the per-IP throttle works, but every 200th connection clears the map, so a client can time reconnects around the cleanup and slip past the throttle.

Fix

Compare elapsed time, reusing the currentTime already computed in the same block:

throttleTracker.values().removeIf(time -> currentTime - time > connectionThrottle);

This is a content-only change to the existing added line in the patch, so it applies cleanly with no rebuildPatches churn.

Fixes #13984.

[gunjanjaswal]

Thanks for the reviews and merge, @electronicboy and @lynxplay!