FwLite push listener doesn't recover after network loss / OS suspend; project-open can crash ('SendCoreAsync … not active')

After network turbulence or an OS-frozen app (Android Doze), FwLite''s SignalR push listener can end up dead-but-cached:

• Opening a project intermittently fails with InvalidOperationException: The 'SendCoreAsync' method cannot be called if the connection is not active (seen in production logs; recurs after connectivity loss followed by a few minutes of idle).
• More commonly the failure is silent: push notifications (auto-download of other users'' changes) stop and don''t recover until a manual sync or app restart. FwLiteWeb has no recovery trigger at all once the initial listener start fails. Likely overlaps the long-standing "auto-download stops" report (internal lb-8mg).

Log analysis of the crash: the cached HubConnection was stuck in Reconnecting (infinite retry, up to 60s between attempts); HTTP connectivity recovered before the next retry tick, so user actions raced ahead of the listener — the crash landed 76 ms after the first successful HTTP request post-outage. The same stuck state explains the silent variant.

Fixed by #2317 (signalr-retry-improvements-v2). Depends on the #2306 fix (#2308) landing first: the listener''s logout detection consumes auth state that was unreliable before it.

Start with the FwLite SignalR push listener and project-open path, then read the fixes in #2317 and its dependency chain through #2306/#2308. Done means the listener recovers after network loss or OS suspend, push notifications resume, and project opening no longer raises the inactive SendCoreAsync error.