webui: actually close the signaling WebSocket

[Rust-KVM]

Requested by 4c · project thread

Summary

Before: #46 did not stop the leak. After a session was taken over, dropped or failed to connect, the signaling WebSocket stayed open, so each Reconnect still added one more established connection on the device.

After: the signaling socket is closed when a session ends (taken over or lost) and on every failed connection attempt (send-offer error, ICE failure, data-channel timeout).

How

gloo-net's Sink::close is a no-op and its WebSocket has no Drop, so #46's sink.close() did nothing. connect() now creates the raw web_sys::WebSocket, wraps it with WebSocket::try_from for the existing stream/sink code, and keeps the raw handle in Peer::signaling_ws. Peer::close() and each error path call close() on that handle. Adds the WebSocket web-sys feature.

Testing

• cargo +nightly fmt --all (CI)
• cargo +nightly clippy --workspace --release --all-targets -- -D warnings (CI)
• cargo +nightly test --workspace (CI)
• Frontend touched: trunk build --locked and wasm32 clippy in crates/rkvm-web (CI)
• Tested on an RK3588 device: two browser tabs took the session from each other and clicked Reconnect repeatedly. Every old signaling WebSocket received a close and only one was open at any time; ESTABLISHED connections on the device's port 80 stayed at 3 after 6 and after 12 takeovers (previously grew from 2 to 10). The app's total fd count still grows by about 3.5 per session; those are non-TCP (UDP sockets / dmabuf) and are tracked separately, not caused by this change.

Checklist

• No #[allow(...)] suppressions and no new unwrap() / expect() in production paths
• Hot paths stay zero-copy; periodic loops use MissedTickBehavior::Delay
• Shared protocol types live in rkvm-proto
• RPC method set changed → not applicable
• Module tree or build invariants changed → not applicable

🤖 Generated with Claude Code

https://claude.ai/code/session_01PbdzY6PfT28goR4BhqdsXR