statement-gossip: Parallelize statement processing and async peer sends

[alexggh]

Problems

• Versi testing showed that when more than 10 peers are connected and sending statements, the time spent in on_statements for decoding, hashing, and deduplication becomes significant and a bottleneck for processing statements. statement-store: process messages from peers in parallel #10892

• When propagating statements, slow peers might block the main loop because send_async_notification was awaited sequentially for each peer. statement-store: make broadcasting multithreaded #10820

High-Level Changes

• Introduce on_statements workers that handle decoding, hashing, and deduplication off the main loop. A configurable number of workers (num_submission_workers) process incoming statement notifications in parallel.
• Keep statement validation workers, the already existed and they handle statement store submission/validation, allowing cryptographic verification to happen concurrently.
• Non-blocking peer sends: Replace sequential awaits on send_async_notification with a FuturesUnordered collection (pending_sends) that queues send futures and polls them from the main loop.

New Flows

Statement Reception:

   Network notification → main loop queues OnStatementsRequest
         → on_statements worker decodes & deduplicates
         → worker sends aggregated ReputationChange back to main loop
         → queues (hash, statement) for validation
         → submission worker validates via statement_store.submit()
         → sends WorkerEvent::StatementResult back to main loop
         → main loop updates peer reputation

Statement Propagation:

    propagate_timeout fires → main loop gets recent statements
         → encodes chunks per peer → creates send futures with timeout
         → pushes to pending_sends (FuturesUnordered)
         → main loop polls pending_sends with highest priority
         → handle_send_result processes completions

Shared State:

The minimal downside is that there is a shared state for this workers, but benchmarks shows updating the shared state is fast and does not contain the bulk of the time

• PeersState: peer known_statements (RwLock)
• PendingState: pending statement→peers map for deduplication (RwLock)

TODO

• Evaluate performance on real world scenarios in versi
• More testing that the minimal one provided.