The alter data step of the new image data refresh process could not finish

[krysal]

Description

The alter_data_batch of the new staging_image_data_refresh DAG causes the Airflow instance to crash. #5145 attempted to fix it by increasing the size of batches, creating fewer tasks to expand, but the result didn't change. This process consumes a lot of the Airflow instance memory, even with the number of active tasks restricted to 2 (#5125), so it needs to be optimized. An alternative is to convert it to an iterative task, similar to how the batched_update operates.

We know the rest of the steps work, given that the staging_audio_data_refresh DAG ran successfully, but the alter process is exclusive to the image table.

Reproduction

Take special care and monitor closely when testing this DAG.

1. Pause the legacy data refresh DAGs once the process finishes
2. Unpause the staging_image_data_refresh and trigger it
3. The UI will start to be inaccessible, so the DAG must be stopped when it is up again or via Airflow CLI.

Additional context

Part of #3925.