log4j-cassandra: failed appender startup leaks the DataStax Cluster, so the JVM never exits

## Description

`CassandraManager.startupInternal()` fails → the `Cluster` is never closed → the JVM never exits, because the DataStax 3.x driver's threads are non-daemon.

`cluster.close()` is only reachable from `shutdownInternal()`, and `AbstractDatabaseManager.shutdown()` skips that when `running == false` (see #4241).

**Version:** 2.x `04c93c1d33` (no 3.x release) · Cassandra 4.1.11, driver 3.11.x

## Logs

```
main DEBUG Shut down CassandraManager Cassandra, all resources released: true
```

Thread dump 8s later — five non-daemon threads, each carrying the configured `clusterName`:

```
"log4j-bench-nio-worker-0" "log4j-bench-nio-worker-1" "log4j-bench-timeouter-0"
"log4j-bench-connection-reaper-0" "log4j-bench-scheduled-task-worker-0"
```

## Reproduction

Point a `Cassandra` appender at a live node with the keyspace **absent**, log one event, call `LogManager.shutdown()`, return from `main`. The JVM hangs.

## Suggested fix

Close the `Cluster` when `startupInternal()` fails — wrap its body in `try`/`catch` that calls `cluster.close()` and rethrows, or make `shutdownInternal()` null-tolerant and let it run irrespective of `running`, per #4241.

Read CassandraManager.startupInternal(), AbstractDatabaseManager.shutdown(), and shutdownInternal() first, then reproduce with a missing keyspace and LogManager.shutdown(). Done means failed startup releases the Cluster and the JVM exits without the five non-daemon driver threads.

Written by the indexing model from the issue text.