feat: DLQ support for processing strategies with batch step

[mj0nez]

Currently, handling of invalid messages with the available DLQ does only work if the strategy has a direct processing chain and does not include batching. Consider the following:

0) consume -> 1) batch -> 2) process in batch -> 3) unbatch -> 4) process message wise -> 5) produce -> 6) commit

All (individual) strategies after the batch step (2 & 4) cannot raise an InvalidMessage exception to trigger the DLQ, because any exception would dump the current batch of messages. Furthermore, if we need the faulty message in the DLQ, the strategy needs some kind of routing producer, which routes valid messages to one and invalid messages to the DLQ topic.

To avoid this, I would like the StreamProcessor to expose it’s handle_invalid_message to the strategy factory, like it does with commit. This would introduce considerably changes to the API but also allow strategies to pass invalid messages directly to the dlq handler, utilizing the configured policy and buffered messages, while avoiding complex routing logic and replicating the DLQ handling.

[mj0nez]

On further thought I think the following changes would be necessary to make this work:

• inject 'handle_invalid_message' to the strategy factory
• adjust the signature of 'create_with_partitions'
• adjust the task strategies to pass 'InvalidMessage' exceptions to the provided handler

Which would move the responsibility of triggering the DLQ-logic from the processor to the strategies, but keep it consistent for batch and "simple" processing.

To put individual messages from batches to the DLQ (step 2 in the example above) the strategy would also need to pass the handle to the function. Thus, another breaking change or we could simply add a new 'RunTaskOnBatch' strategy which covers that.