Add OpenMLDB multi-table source support

The OpenMLDB source accepted only one root-level SQL query and relied on SDK input-schema discovery. Queries with different result schemas required separate source configurations, while validation also exposed null, row-reuse, connection-ownership, and cluster-query correctness issues.

Validates unique output identities and configured result schemas, matches aliased columns by case-sensitive names, preserves nullable values, copies reused row arrays, keeps shared connections alive across readers, and directs cluster reads to online rows instead of the offline-job path. Existing root-level SQL remains supported.

• Allows a single source configuration to emit records from multiple OpenMLDB queries with distinct schemas and table identities.
• Fails mismatched names, counts, and types instead of silently mapping incompatible results, and reports completion only after every query succeeds, including empty results.
• Corrects SQL NULL handling and reader isolation while preserving existing connector options, defaults, registration, and SDK version.
• This does not add parallelism, CDC, exactly-once delivery, or a consistent snapshot across tables. Streaming repeats configured queries and can emit duplicates.

• Twenty-nine connector unit tests passed on Java 8 and Java 11; the Java 11 dependency reactor passed 506 tests with no failures, errors, or skips.
• Five opt-in real-server integration tests passed against OpenMLDB 0.6.3 in Java 8/11 and standalone/cluster combinations: 20 executions verifying 1,101 complete unique rows, nullable values, schemas, aliases, empty results, isolation, failures, and completion.
• GitHub Build passed, maintainer davidzollo approved with 'Good job,' and the verified merge commit credits Goutam Adwant.
• The real-server tests require Linux amd64 and a disposable server. Zeta, Flink, and Spark engine E2E validation was not reported and is not implied by connector-level testing.