feat(kafka): add topic pattern subscription support

[touhou09]

Summary

Add support for subscribing to multiple Kafka topics using regex patterns.
This enables scenarios where a single Arroyo pipeline can consume from multiple topics matching a pattern (e.g., ^logs-.* or ^edge[0-9]+_raw$).

Motivation

Currently, Arroyo's Kafka source connector only supports subscribing to a single explicit topic. In many real-world scenarios (e.g., multi-tenant systems, edge device data collection), data is partitioned across multiple topics with a common naming pattern. This PR adds the ability to subscribe to all topics matching a regex pattern using Kafka's native pattern subscription feature.

Changes

• Add topic_pattern option to Kafka source table configuration
• Implement pattern-based subscription using rdkafka's subscribe() method
• Update KafkaState to track offsets by (topic, partition) key for multi-topic support
• Add validation: topic and topic_pattern are mutually exclusive
• Add validation: topic_pattern only valid for source tables, not sinks
• Update existing tests to work with new field structure

Usage Example

CREATE TABLE multi_topic_source (
    data TEXT
) WITH (
    connector = 'kafka',
    type = 'source',
    topic_pattern = '^logs-.*',
    format = 'json',
    bootstrap_servers = 'kafka:9092',
    'source.offset' = 'earliest'
);

SELECT * FROM multi_topic_source;

Behavior

When using topic_pattern:

• Kafka handles partition assignment via consumer group protocol
• New topics matching the pattern are automatically discovered
• Rebalancing happens automatically when topics are added/removed
• The topic metadata field reflects the actual topic each message came from

Breaking Changes

None. The topic field remains the default and recommended option for single-topic subscriptions.

Test Plan

• Updated existing unit tests to use new field structure (Option<String> for topic)
• Integration tests with pattern subscription (requires Kafka cluster)
• Manual testing pending

Related Issues

This addresses the need for multi-topic subscription in stream processing pipelines.