Tracking issue about todo in distributed wal

[Rachelint]

Description
There are some enhancements in wal needing to do in the futures, just mark them now.

Proposal

On message

• Now manifest based on wal need to scan twice from storage while doing snapshot. Add a cache for this may be better.
• In wal's message queue implementation, we now sync the snapshot to remote in every flushing. Indeed, we should just need to do the sync work while flushing after dropping table.
• Now, the desigin of wal on message queue is based on the thread model of one writer for one table, maybe we should remove this assumption.
• Define some iterators with more explicit semantics.
• Now we init region(heavy work with network IO) even if it is just created, we should change MessageQueue to know whether it is a new region.
• We should disable the ttl of topic in kafka, but it is not supported in rskafka now. It is important so I plan to use another crate to finish it temporarily.

On OBKV

• Rename the table name prefix.
• Change the meta table's sharding key to table id.

Manifest(on wal)

• Use the table id as region id.

Additional context