Continue gathering acks for synchro transactions after restart

[sergepetrenko]

Now if a restart happens while some synchronous transactions are in flight, these transactions are left hanging until manual intervention with box.ctl.clear_synchro_queue or a new leader being elected, in case of raft.

Instead, master should gather acks for pending synchronous transactions after restarting, if it sees that there was no intervention into its limbo (noone else confirmed the transactions).

[sergepetrenko]

Blocked by #5445

[sergepetrenko]

We shouldn't do this. Now the master's transaction limbo is frozen until it either receives a promote from remote instance or box.ctl.promote() is called manually on this instance. The recovered node should never confirm anything without consulting a quorum first.