Worker startup race: NetworkActor drains gossip before StateActor is ready

[intendednull]

Problem

In crates/worker/src/runtime.rs:36-60, actors are spawned sequentially without synchronization:

let state_addr = system.spawn(StateActor { role });     // Line 39
system.spawn(NetworkActor::new(state_addr, ...));        // Line 41

NetworkActor::started() spawns a background task that immediately begins draining TopicEvents and forwarding EventMsg to the StateActor. If the gossip topic has buffered messages, they arrive at the StateActor before its started() hook completes.

There is no startup barrier — no mechanism to ensure all actors are initialized before processing begins.

Impact

• Events arriving during startup may hit an uninitialized WorkerRole
• Role-specific invariant checks may panic on null/uninitialized state
• Race is invisible in tests (in-memory networks have no pre-buffered messages)

Suggested fix

Options:

1. Add a Ready message that NetworkActor sends after all actors confirm initialization
2. Buffer events in NetworkActor until a StartProcessing signal is received
3. Use a tokio::sync::Barrier across actor startup

Location

• crates/worker/src/runtime.rs:36-60
• crates/worker/src/actors/network.rs:46-59 (background task spawned in started())

References

Found during deep implementation audit (pass 2)