fix: resolve consumer deadlock on partition rebalance when paused (#681)

[DmytroLiulchuk]

Description
Fixes #681

When a consumer has all partitions paused, KafkaFlow starts a heartbeat background task that calls Consume(1000) in a loop to maintain group membership. If a partition rebalance occurs during this Consume() call, librdkafka invokes the revoke callback (configured via Confluent.Kafka's SetPartitionsRevokedHandler) synchronously on the same thread (documented behavior). The handler calls ConsumerFlowManager.Stop() → StopHeartbeat(), which previously awaited the heartbeat task synchronously via _heartbeatTask.GetAwaiter().GetResult(). Since this executes on the heartbeat thread itself, the thread waits for its own completion — a deadlock.

Fix: Remove the synchronous wait in StopHeartbeat(). After Cancel(), the heartbeat task exits naturally on the next loop iteration (while (!IsCancellationRequested) → false), releases the semaphore in finally, and the feeder unblocks.

How Has This Been Tested?

• Consumer A pauses all partitions (triggers heartbeat)
• Consumer B joins same group (triggers rebalance during heartbeat's Consume(1000))
• Without fix: consumer group deadlocks permanently
• With fix: consumer group recovers and processes new messages

Checklist
 [+] My code follows the style guidelines of this project
 [+] I have performed a self-review of my own code
 [-] I have added tests to cover my changes
 [-] I have made corresponding changes to the documentation