Scale write to a single replicated aggregate

[krasserm]

This can be achieved by allowing an EventsourceActor (= aggregate) to consume from multiple logs and only write to one of these logs. Aggregate replicas with the same aggregate id write to different logs but consume from all of them, so that distributed writes to the same aggregate id don't need to be merged into a single log. For example, with three (optionally replicated) logs, A, B and C and three aggregate replicas r1, r2 and r3:

• r1 writes to A and reads from A, B and C
• r2 writes to B and reads from A, B and C
• r3 writes to C and reads from A, B and C

This means that replicas ri need to be recovered from three logs (instead of one) and deterministic (= totally ordered) replay is not possible any more. However, by re-ordering concurrently replayed events based on their vector timestamps, causal ordering can be achieved.

[krasserm]

This is a scalability extension to #26 and #36 that preserves causal ordering as #13 alone does not.

[krasserm]

Obsolete with #66 therefore closing this issue.