[network] connection_events() is a placeholder that never yields

[intendednull]

Parent: #108

Problem

crates/network/src/iroh.rs:345-349:

async fn connection_events(&self) -> ConnectionEventStream {
    // Placeholder: return a stream that never yields.
    // Full implementation would monitor endpoint relay and direct connection state.
    Box::pin(futures_lite::stream::pending())
}

crates/network/src/mem.rs:378-381 has the same pending stream.

Anything awaiting connection_events() hangs forever. Callers assume they've subscribed to relay/peer connect/disconnect events — they haven't.

Fix

Two options:

Option A: implement it properly. Hook into iroh's underlying connection-state events and emit ConnectionEvent::RelayUp, RelayDown, PeerConnected, PeerDisconnected. For MemNetwork, emit synthetic events when MemHub subscribes/unsubscribes a peer. This is the right long-term answer.

Option B: delete the method and update the trait. If no in-tree caller actually uses the stream (grep for connection_events across crates), just remove the trait method and each impl. Add a TODO(#TBD) if we expect to need it later.

Start by doing a workspace-wide grep for connection_events callers. If there are any, option A. If not, option B.

Test

If implementing: a MemNetwork test that subscribes two peers, connects one via the hub, and asserts a PeerConnected event is emitted within a bounded time.

If deleting: just make sure nothing breaks and document the rationale in the commit message.

[intendednull]

Resolved in 7bac12e ("network: remove ConnectionEvent / connection_events stub"). Option B from the issue was taken — the placeholder method was removed from the Network trait and both impls since no in-tree caller used the stream. See crates/network/src/traits.rs:144-146 for a TODO(#119) marker preserved for future re-addition. The hanging-stream defect no longer exists.