[Improvement] register only the required catalogs, not all of them for spark-connector

### What would you like to be improved?

All catalogs are registered during the initialization even if a spark sql does not use them.

The action may result in:
1. The list operation slows down if there are many catalogs.
2. Registering too many catalogs to memory increases memory pressure

```
com.datastrato.gravitino.spark.connector.plugin.GravitinoDriverPlugin#init

public Map init(SparkContext sc, PluginContext pluginContext) {
...
Set catalogs = catalogManager.listCatalogs();
registerGravitinoCatalogs(conf, catalogs);
registerSqlExtensions();
return Collections.emptyMap();
}
```

### How should we improve?

_No response_

Start at com.datastrato.gravitino.spark.connector.plugin.GravitinoDriverPlugin#init and inspect how catalogManager.listCatalogs() feeds registerGravitinoCatalogs. Determine how the connector can identify catalogs required by Spark SQL, then verify that initialization registers only those catalogs and avoids the listed performance and memory costs.

Written by the indexing model from the issue text.