fix(cbf): shut down chain source

[febyeji]

• Implemented CbfChainSource::stop() by shutting down the active Kyoto requester.
• Marked the CBF runtime status as stopped when the Kyoto node exits cleanly.
• Routed generic ChainSource::stop() calls to the CBF backend.

[randomlogin]

I think we can also add changes you introduced here: 0181500, reordering the stop and wait_on_background_tasks.

[randomlogin]

There is a corner case with restarts: when the node crashed and we wait for the backoff time to tick, if we call stop() it will be ignored and the node rebuilds.

We can solve it as it was previously: https://github.com/febyeji/ldk-node/blob/add-cbf-chain-source/src/chain/cbf.rs#L403

Perhaps can be fixed in the following way (claude-generated):

// Publish the new requester only if stop() didn't fire during the
// backoff sleep. Holding the lock across both the check and the
// assignment makes this atomic w.r.t. stop(): either stop() ran
// first (we observe `Stopped` and bail), or we publish `Started`
// and a later stop() shuts the new node down once it's running.
{
        let mut status = restart_status.lock().expect("lock");
        if matches!(*status, CbfRuntimeStatus::Stopped) {
                log_info!(
                        restart_logger,
                        "CBF restart aborted: stop() called during backoff.",
                );
                break;
        }
        *status = CbfRuntimeStatus::Started { requester: new_requester };
}

current_node = new_node;
current_info_rx = new_info_rx;
current_warn_rx = new_warn_rx;
current_event_rx = new_event_rx;

[randomlogin]

LGTM