Fix band check bug in split_raster and typo fixes

[abhayrajjais01]

• Fix bug in preprocess.py where band count check used shape[2] (width) instead of shape[0] (channels) after transposing to channels-first format
• Fix typo 'prediciton' -> 'prediction' in evaluate.py docstring
• Fix typo 'prediciton' -> 'prediction' in predict.py comment

Description

band check was reading width instead of channels after transpose

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 87.16%. Comparing base (3146b96) to head (b391811).
⚠️ Report is 17 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1301      +/-   ##
==========================================
- Coverage   87.89%   87.16%   -0.74%     
==========================================
  Files          20       21       +1     
  Lines        2776     2851      +75     
==========================================
+ Hits         2440     2485      +45     
- Misses        336      366      +30     

Flag	Coverage Δ
unittests	87.16% <100.00%> (-0.74%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[henrykironde]

@abhayrajjais01, thank you for the PR. We need a bit more information to properly document this patch and the findings. Could you please let us know how you discovered this bug, specifically which scripts or functions you were running at the time?

[abhayrajjais01]

I found this while reading the split_raster function to understand how numpy images are validated. After transpose(2, 0, 1) converts (H, W, C) to (C, H, W), line 210 reads shape[2] which is now the width dimension. For a 400×400×3 image, bands = 400, the warning fires incorrectly, and numpy_image[:3, :, :] selects 3 rows rather than 3 channels. I also noticed that if shape[2] is not in [3, 4] (like 5-band input), the image is not transposed, but [:3, :, :] still runs on a channels-last array
would you like me to address this in a follow-up?