Fixed incorrect state with resume then reconnect (#595)
When a resume sequence fails, a full reconnect is attempted. There were
a few issues there with that sequence
- We did not always fire EngineEvent.Restarting, so Room missed tearing down existing participants
- With selective subscriptions, when existing `isDesired` isn't cleared, it will not send a subscribe request when requested
- New tracks were not republished successfully when reconnected (due to sender not being reset early enough)
