Columnar: Review cluster shutdown behavior

The RFC has the requirement of handling shutdown gracefully - at the moment the core only has support for ShutDownNow. Review that behavior, and explore the possibility of adding support for ShutDown with a grace period.

ShutDown(gracePeriod: Duration)

• Shuts down the topology tracker / disconnects from all KV nodes.

• Future requests fail with an idiomatic “cluster is already shut down” error.

• Waits for all in-flight requests to complete or for the grace period to elapse, whichever comes first.

• Any remaining in-flight requests fail with an idiomatic “operation canceled due to shutdown” error.

ShutDownNow()

• Equivalent to ShutDown(Duration.ZERO)