Fix pre/post-training evaluation to use same batch in nn_tutorial

[patrocinio]

The tutorial was comparing loss on different batches:

• Pre-training: evaluated on first 64 instances (batch 0)
• Post-training: evaluated on last batch from training loop

This made the comparison misleading as it wasn't measuring improvement on the same data.

Changes:

• Save the initial batch (xb_initial, yb_initial) after first evaluation
• Use the saved initial batch for post-training evaluation
• Added clarifying comment about fair comparison
• Now both evaluations use the same data (first 64 training instances)

This provides an accurate before/after comparison showing the model's improvement on the same batch of data.

Fixes #3666

Description

Checklist

• [ ] The issue that is being fixed is referred in the description (see above "Fixes #ISSUE_NUMBER")
• [ ] Only one issue is addressed in this pull request
• [ ] Labels from the issue that this PR is fixing are added to this pull request
• [ ] No unnecessary issues are included into this pull request.