Decouple fasterq-dump from prefetch to make processing of large datasets faster

[misialq]

This enhancement request requires some more evaluation.

In case of large datasets it takes very long for fasterq-dump (blue line in the screenshot below) to process the sequences fetched with prefetch (simply due to the size; red line). As these two actions (prefetch -> fasterq-dump) are currently chained within one function, nothing else can be processed or downloaded before every single sample is done.

This could be significantly improved by letting prefetch grab everything first and while some data is already available fasterq-dump could start processing that (similarly to how we do it later for the post-processing steps).