Absolutely, I’d agree that being prepared and responding correctly to a disruption is the way forward here.
A connection could drop for a number of reasons - network disruption, data center change / failover / load balancing, code deployment on our side, etc.
Note that the public streaming API does not have any fallback option, but the commercial Gnip offerings do offer the option of redundant connections and also backfill for missed data.