Fix execute hang in remote mode

[andrii-i]

Fixes #87

Problem

nb execute in remote mode can hang indefinitely. The tokio::select! in the kernel message loop waits for WebSocket messages and Y.js updates but has no timeout arm. The per-cell deadline is only applied after Status(Idle) arrives. If that message is missed (race condition observed after nb output clear + nb execute in quick succession), the loop blocks forever.

A secondary issue: when the kernel WebSocket closes, recv_message() returns Ok(None). The if let Some(msg) pattern silently skips None, re-enters the loop, and spins at 100% CPU since a closed socket returns None immediately on every read.

Changes

Added timeout arm to tokio::select!. sleep_until(deadline) now serves as a select branch so the loop always exits within the configured per-cell timeout, regardless of whether Status(Idle) arrives. Matches the same pattern the local executor already uses (timeout_at wrapping each recv).

Fixed WebSocket close handling. Changed if let Some(msg) to match with an explicit None => break arm. On WS close the loop now exits into the fallback path instead of spinning.

Improved fallback path to collect outputs. Previously the fallback returned a blind ExecutionResult::success. Now it reads any remaining outputs from the Y.js document and uses from_outputs to detect errors, so a timed-out execution that produced an error still reports as failed.

Extracted ExecutionResult::from_outputs helper. Moved the error-detection scan (previously inlined in the happy path) into a reusable method on ExecutionResult. Single-pass find_map over outputs replaces the prior any() + find_map() double scan.