* Don't kick the user on network errors * Track online status for RQ * Use health endpoint * Update test with new behavior * Only poll while offline * Handle races between the check and network events * Reduce the poll kickoff interval * Don't cache partially fetched pinned feeds This isn't a new issue but it's more prominent with the offline handling. We're currently silently caching pinned infos that failed to fetch. This avoids showing a big spinner on failure but it also kills all feeds which is very confusing. If the request to get feed gens fails, let's fail the whole query. Then it can be retried.