Provide better support for transient state machine implementation

[drmingdrmer]

An application might implement a pure in-memory state machine, which will be discarded when server is shutdown, and will be rebuilt from the last snapshot the next time server starting up.

In such a scenario, the state of the state machine falls back to a previous state after restart. And openraft should help the app to rebuild the state machine to the latest state.

TODO:

• Upon startup, a node that is leader should restore local log progress and re-apply log entries if it is single node cluster #883
• Consider informing the storage to persist the last_applied_log_id, even when applying log entry to a pure in-memory state machine; And reload the persisted last_applied_log_id when starting up;

[github-actions]

👋 Thanks for opening this issue!

Get help or engage by:

• /help : to print help messages.
• /assignme : to assign this issue to you.