Evaluation code needs to check in case there are inconsistencies in dimension order for template/target vs prediction in ALPS domain

[RosieEade]

The current code in evaluation/* breaks if the dimensions of the target and prediction have a different order.

See issue #43

For the original version of training and submission code, the target and templates files for the ALPS domain have dimensions (time, y, x), but the output predictions had dimensions (time, x, y). This means that difference computations fail to align the dimensions correctly.

I am looking into how to check for this issue and apply a correction to the data.