perf(l1): use secondary instances for reading in RocksDB

[Oppen]

WIP: we want to see if using secondary instances improves read performance, and maybe later on compaction.
The initial approach is to make it just work to see possible issues, understand how it works, etc, so it tries to catch up at every write step and keeps both instances inside a single Engine instance.
We'll probably want to handle it different before considering the feature ready.
First, we might want to split it so we can make multiple secondary instances, and be more flexible about catch-up frequency.
For example, the RPC probably can live slightly out of date, while block production tends to need the latest version and the primary instance might be better for all operations there. The P2P handler is likely to need both so it can respond to requests from the secondary while having access to the primary for when it receives responses to its own queries.