API design for inputs

The question is about what is the OSnet library expecting as an input dataset:

- should only the space/time coordinates given ? hence SLA, topo, etc filled internally before prediction
- should all of the 12 inputs be provided ? (no internal help)
- a mix of the above ?

and then, technically:
- how to identify variables within the input datasets ? (probably following cf convention)
- what about the grid ? should we broadcast internally all inputs against each others ? or let users manage this upstream of the prediction ?

We should also not try to cover all possibilities in the 1st release, but rather provide a starting point and wait for user feedbacks