fix: Disconnect peers on handshake failure

[diegomrsantos]

Issue Addressed

#438

Proposed Changes

• Add explicit peer disconnection: When a handshake fails (e.g., due to network mismatch), the network layer now explicitly disconnects the peer via disconnect_peer_id()
• Improve disconnection timing: Instead of relying on keep-alive timeouts (~10 seconds), peers are disconnected immediately upon handshake failure
• Preserve existing test: The handshake test continues to validate failure detection; disconnection behavior will be naturally tested through integration tests that use the full network stack
• Add testing best practices: Updated CLAUDE.md with guidance about not simulating production behavior in tests

Technical Details

The change is made in Network::handle_handshake_result() where handshake failures are processed. When a handshake fails, we now call self.swarm.disconnect_peer_id(peer_id) to immediately disconnect the peer, rather than waiting for keep-alive timeouts.

This ensures faster and more deterministic peer disconnection on handshake failures, improving network behavior and reducing connection overhead.

[Copilot]

Pull Request Overview

This PR improves the mismatched_networks_handshake_failed test by adding verification that peers are properly disconnected when handshake fails due to network mismatch. The existing disconnection functionality was already working correctly - this change adds test coverage to verify the behavior.

• Adds tracking of connection close events for both local and remote peers
• Extends the test loop condition to wait for both handshake failures and disconnections
• Adds explicit assertions to verify both handshake failures and disconnections occurred

[dknopik]

I think this works because at some point there is a keepalive timeout as nothing is being sent anymore:

2025-07-21T09:33:38.548712Z DEBUG Swarm::poll: libp2p_swarm: Connection closed with error KeepAliveTimeout: Connected { endpoint: Listener { local_addr: /memory/4092149446297013635, send_back_addr: /memory/12899751130364311786 }, peer_id: PeerId("12D3KooWB1vGoK219R7Dnw7YUi4EcViLPWqjanY8F9P4TF27xL7u") } total_peers=0
2025-07-21T09:33:38.548789Z TRACE libp2p_swarm_test: event=ConnectionClosed { peer_id: PeerId("12D3KooWB1vGoK219R7Dnw7YUi4EcViLPWqjanY8F9P4TF27xL7u"), connection_id: ConnectionId(2), endpoint: Listener { local_addr: /memory/4092149446297013635, send_back_addr: /memory/12899751130364311786 }, num_established: 0, cause: Some(KeepAliveTimeout) }
2025-07-21T09:33:38.548814Z TRACE async_io::block_on: async_io::driver: sleep until notification

The test takes 10 seconds.

So I think #438 is actually valid.

We should modify this test to ensure the disconnection reason is correct (if possible), or ensure that the disconnection is soon after handshake failure.

[dknopik]

Nice, thanks!

The base branch was changed.

[dknopik]

Sorry, can you please rebase this from unstable to release-v0.3.0?

[diegomrsantos]

It seems to be good to be merged into release-v0.3.0

[dknopik]

Currently, other changes are in this PR: 8f9cc96 and 17b8f05

[diegomrsantos]

Currently, other changes are in this PR: 8f9cc96 and 17b8f05

done

[dknopik]

LGTM, thank you!