Share more logic between different DB implementations.

[ruseinov]

Currently the implementations of Blockstore, PersistentStore and other interfaces are completely independent for MemoryDB and ParityDB. Which means that if we test e.g. GC on MemoryDB - it's not guaranteed to have the same behaviour on ParityDB.

Perhaps a different approach should be considered, where more of the DB logic is shared.

For example the persistent storage fallback:

    fn get(&self, k: &Cid) -> anyhow::Result<Option<Vec<u8>>> {
        Ok(self
            .blockchain_db
            .read()
            .get(&k.to_bytes())
            .cloned()
            .or(self
                .blockchain_persistent_db
                .read()
                .get(&k.to_bytes())
                .cloned()))
    }

Could be completely generic across multiple databases. One way to achieve that is introducing a forest-specific (or chainsafe-specific) trait that all the database in question would implement. Then the implementation for many of the other traits could be based on the exposed interface as opposed to being database-specific.

[ruseinov]

Related to e.g. #4959