Summary

Error recovery/rollback operations and background stream handlers were silently discarding failures via bare let _ = or missing error branches. When these paths fail, the system loses its ability to restore consistent state—and there's no observable signal that something went wrong.

This PR adds diagnostic log entries (append_diagnostic_log) to all critical silent-failure sites without changing control flow or error propagation semantics. The pattern used:

if let Err(err) = fallible_recovery_op() {
    let _ = crate::diagnostic_log::append_diagnostic_log(
        "module.event_name",
        serde_json::json!({ "error": err.to_string(), ... }),
    );
}

Key changes:

• relay_switch.rs — Log when store.save() fails during rollback after a failed apply_selected_relay_profile, and when settings load falls back to defaults
• relay_config.rs — Log failed restore_optional_file calls for both auth and config paths during write-error rollback
• launcher.rs — Log stream chunk read errors (previously break with no trace), helper connection errors (fire-and-forget task), and failure-status persistence errors
• watcher.rs — Log command.spawn() failure when relaunching on Windows
• provider_sync.rs — Log restore_session_changes failure during apply-error rollback

Link to Devin session: https://app.devin.ai/sessions/00a57b2e950a4c45b346a605e295aad6
Requested by: @ManhND226677