Replay from Cassandra stops prematurely if there are large sequence number gaps covering empty partitions

[volkerstampa]

After disaster recovery the local sequence is potentially adjusted to ensure that it is larger than the local log's time (see also commit for #286). If Cassandra is used as backend and the gap introduced through this adjustment covers empty partitions, replays from this event log stop prematurely as they do to proceed after the first empty partition.

[volkerstampa]

This fix had to be reverted with #407