[3.1.9 Backport] Detect and handle _sync:seq rollback in sequence allocator

The sequence allocator assumes that the next allocated sequence is always going to be larger than the previously seen sequence.  There are a few scenarios where _sync:seq could be rolled back on the server:

• vbucket failover/rollback
• incorrect migration of data into the bucket (XDCR, backup/restore)
  If _sync:seq is rolled back while Sync Gateway is running, the sequence allocator has the ability to detect this and restore _sync:seq to a value greater than the last allocated sequence on the node.

The solution needs to consider the case where multiple nodes detect this rollback concurrently.