Allow to split EKO computation for a dataset

[felixhekhorn]

While the computation of a single 4D EKO is reasonably fast, some datasets have many scales which makes them slow again. This is becoming a bottle neck on clusters, where a time limit of 24h seems to have risen.

Starting from NNPDF/eko#296 EKO provides some support to split the computation into several pieces. Note NNPDF/eko#399

@scarlehoff is wishing for

Pineko-wise it would be enough being able to continue an eko where we left it
So I get an operator card and an eko.tar file and instead of refusing to run or rerunning it looks inside and continues

this should be doable

[alecandido]

It is, but you can even be more ambitious (if needed) since the recipes mechanism was made on purpose for that. So, the recipes generation is instantaneous, after that, you can decide which ones (or simply how many) you want to compute right now, or somewhere. Then, you will only consume the recipes you have computed (though marking them as done is a more or less manual operation).