Split up StatusDeque into different modules
* LastIdQueue tracks last_ids
* StatusCache keeps track of signature statuses
* StatusCache stores success as a bit in a bloom filter
* Overhead for 1m Ok transactions is 4mb in memory
* Less concurrency between the objects, last_id and status_cache are read and written to at different points in the pipeline
* Each object has its own strategy for merging into the root checkpoint
This trait is for bloom, not crds.
Slightly better would be to put it in the SDK so that the trait
implementations could go into hash and pubkey, but if we don't
want compatibility constraints, this is the next best thing.
* Cluster Replicated Data Store
Separate the data storage and merge strategy from the network IO boundary.
Implement an eager push overlay for transporting recent messages.
Simulation shows fast convergence with 20k nodes.