Client fails to auto-restart a stream if last message was a snapshot marker

Scenario:

1. Client receives seqno 1.
2. Client receives snapshot marker for snapshot [2-2]
3. Channel disconnects (maybe Couchbase node is restarted) and the client attempts to auto-reconnect.
4. Reconnect fails, because the client tries to create a stream offset with seqno 1, which is not within the snapshot [2-2]

If this happens, the DCP client stops streaming from the affected partition.

Possible fix is to track pending snapshots separately, and only promote them to "current" after the client's seqno advances to within the pending snapshot.