Persistent multi-process Pool

### Feature request

Running .map and filter functions with `num_procs` consecutively instantiates several multiprocessing pools iteratively.

As instantiating a Pool is very resource intensive it can be a bottleneck to performing iteratively filtering.

My ideas:
1. There should be an option to declare `persistent_workers` similar to pytorch DataLoader. Downside would be that would be complex to determine the correct resource allocation and deallocation of the pool. i.e. the dataset can outlive the utility of the pool.
2. Provide a pool as an argument. Downside would be the expertise required by the user. Upside, is that there is better resource management.

### Motivation

Is really slow to iteratively perform map and filter operations on a dataset.

### Your contribution

If approved I could integrate it. I would need to know what method would be most suitable to implement from the two options above.