Save results immediately

[PierreGtch]

Closes #419

Instead of waiting for the evaluation of a whole dataset to finish before saving the results, we could save them as soon as they are computed. Beneficial in case the evaluation fails in the middle.

Pending questions:

• [ ] Can that create issues in case of parallel access to the hdf5 file?
• [ ] What is the computational overhead due to accessing the hdf5 more often?

[PierreGtch]

@bruAristimunha the code change should be quite simple I think (see my commits) but I don't have time now to test the parallel case and benchmark in the standard case.

[bruAristimunha]

Okay, but now we lost all the results in the tutorials with this modification. We will need to debug this issue more to get it all green.

[bruAristimunha]

I wonder if you could focus a little more on this problem in the CI, and then we could create a more specific test to work on the parallelization and overhead access issue. @PierreGtch

[PierreGtch]

I see it fails while testing the benchmark function. By default, benchmark uses -1 jobs. I think it might already be an issue with the parallel access to the file...

When I tested in local, the tests/evaluations.py were working fine