Gracefully stop reth coordinator tasks

[ByteYue]

Summary

• replace the coordinator watchdog's direct process::exit path with an awaited Result-returning shutdown path
• broadcast shutdown when a coordinator task exits, then wait briefly for the remaining coordinator tasks to stop
• join the reth runner thread before returning a non-zero exit for coordinator failures

Validation

• RUSTFLAGS="--cfg tokio_unstable" cargo check -p gravity_node

Related audit context: Galxe/gravity-audit#11