T225578 implemented a queue of events for EventLogging. It also flushes that queue on pagehide. However, unload happens after pagehide, and there's existing logging being performed in unload handlers that's now suppressed by the new queue behavior.
(Arguably, for any browser which supports pagehide we shouldn't be doing logging from unload, given its general unreliability. But not-breaking the current cases seems simpler than finding and rewriting existing logging handlers.)