[Nodes] Add pre-batch option to parallel mapper

[andrewkho]

🚀 The feature

Add a pre-batch option to ParallelMapper to improve performance

Motivation, pitch

Throughput can often be improved by doing pre-batching of inputs, lowering queue overheads. Add an option (default off?) to pre-batch for improved throughput. Maybe we could even tune-this dynamically by measuring queue overhead? Need to be careful about memory overhead so maybe best to leave this off by default

Alternatives

No response

Additional context

No response