Kafka 4.x Queue Semantics support in Kafka Ingestion

The Shared Group API (KIP-932) Kafka handles partition assignment automatically and consumer group coordination can be simplified, hence upgrading to Kafka Share Consumer API can have significant performance improvements.

### 1. Use Case: Breaking the Partition Bottleneck

Currently, Druid ingestion is limited by a **1:1 mapping** between Kafka partitions and Druid tasks.

* **The Problem:** High-volume "hot partitions" create bottlenecks, limiting ingestion scalability.
* **The Solution:** Leverage **Kafka Share Groups**, enabling multiple Druid tasks to consume from the same partition, treating Kafka as a distributed queue.

### 2. Importance & Benefits

By adopting architectural patterns from **Flink FLIP-573**, we achieve:

* **Horizontal Scalability:** Scale Druid tasks independently of Kafka partition count.
* **Instant Scaling:** No costly rebalancing. Adding a task is fast; the Kafka broker handles record assignment.
* **Improved Fault Tolerance:** Unacknowledged records are automatically redelivered to other tasks upon task failure.

The issue concerns Druid Kafka Ingestion and the Kafka Shared Group API from KIP-932, with Flink FLIP-573 cited as an architectural reference. Start by reviewing those APIs and the existing Kafka ingestion design. Done would mean supporting multiple Druid tasks on a partition, scaling without costly rebalancing, and redelivering unacknowledged records after task failure.

Written by the indexing model from the issue text.