Design a custom consensus protocol for a distributed database cluster operating in an asynchronous network environment with high latency (500ms+). The system requires linearizability for writes but can tolerate eventual consistency for reads. Describe the state machine replication process, how leader election is handled during network partitions, and the recovery mechanism for crashed nodes. Compare your solution's trade-offs regarding availability and partition tolerance against the Raft or Paxos algorithms.