global_forecast: solve evaluation inconsistencies for batch size different from 1

[iluise]

When batch_size in evaluator.py is different from 1, then we see that:

1) when running multi-dates global_forecasts only the first lead time is saved. 2) data reconstruction for predictions (not target!) is mis-reconstructed as shown below:

Temporary solution: hardcode batch_size = 1 in evaluator.py

[clessig]

Is only the first lead time or the first date saved? What does it tell us that target is reconstructed correctly?