aws_dynamodb_cdc: Support namespace or client_id for checkpoint isolation

Context: In our development environment, multiple engineers frequently run the same Redpanda Connect pipeline against a shared DynamoDB table (CDC stream) for testing. Currently, all instances point to the same checkpoint_table.

Problem: Checkpoints in the DynamoDB table are currently keyed only by (StreamArn, ShardID). There is no concept of a "Consumer Group" or logical scope. When multiple developers run the pipeline simultaneously:

They contend for the same checkpoint rows. Sequence numbers are overwritten by different consumers. This leads to skipped events, duplicates, and inconsistent state across the dev team.

While creating separate checkpoint tables per developer is a workaround, it creates significant infrastructure sprawl and is operationally difficult to manage at scale.

Ask: Could the aws_dynamodb_cdc input be updated to support an optional namespace, consumer_group, or similar identifier?

The goal is to allow multiple independent logical readers to share a single physical checkpoint table without colliding. This would bring the DynamoDB CDC input in line with the "one store, many logical readers" pattern and significantly improve the developer experience by reducing infrastructure sprawl.

Start by locating the aws_dynamodb_cdc input and the code that constructs checkpoint keys from StreamArn and ShardID. Read the surrounding configuration and tests to determine how an optional logical scope should be represented; done means independent readers can share one checkpoint table without overwriting one another, with coverage for isolated checkpoints.