[Prototype] Simplify PipelineRL

[jlamypoirier]

✨ Description

• Drop the alternative implementations of streaming datasets, keep only the consumer-group one.
• Remove most configuration options for redis, as configurable message names are likely to bring confusion and could be hard to synchronize with PipelineRL. Use hard-coded values instead.
• Drop the new dataset structures, since the existing ones are good enough for streaming datasets. (though we need to overlook some details like sampling configs and sample indexes not making sense)