Improve benchmarking reporting (epic)

[eftsung]

This work aims to improve how the performance benchmarking is reported.

We observe and propose the following:

• Occasionally, we get some strange negative value for iterations per second, i.e.

2025-03-08 14:43:23.01 Epoch 1/2, Step 300/1563, Loss: 6.8957, -160.7 Images/sec

--> Proposal: Find and fix the cause for this or at the very least, filter these results out.

• The iterations per second value for each epoch and overall is displayed only to the nearest integer.

Mean Images per second in this training epoch: 414

and

Total mean images per second in training: 413

• For iterations per second, mean is the only stat displayed.
  -> Proposal: Add other stats: total points, medium, standard deviation. These could give the user greater insight into the performance of the model.

• For slowly running models, we may want to terminate the training/evaluation early.
  --> Proposal: Add maximum number of steps option