Add Neural Network Cleanup

[yoursanonymous]

##Summary
This PR addresses TODOs in the codebase by implementing proper memory cleanup for the neural network component and documenting magic numbers used for buffer padding.

[joseph-yiu]

Please separate this into multiple patches. The NN cleanup need to be linked to #81. But I can't do that when your PR is combining multiple changes. Please separate the patches into 3 pull requests:

(1) The new file .github/workflow/build.yml duplicated with #79
(2) The NN cleanup
(3) Documentation improvement
In future, please also provide the GitHub issue that you are trying to address in the PR description. It is okay here now as there are only a few issues and PRs. But in bigger projects it can be difficult to see the connections.
Thanks and regards,
Joseph

[yoursanonymous]

Thanks a lot for the detailed feedback, Joseph really appreciate it.

I will split this PR into multiple pull requests as suggested:

-A PR for the NN cleanup (linked to #84)

-A separate PR for the documentation improvements

-I will drop the duplicated CI workflow change since it’s already covered in #79

Thanks again for pointing this out and for the guidance.

[joseph-yiu]

Another suggestion, if you don't mind:

Instead of calling the new function th_nn_deinit(), please use the name th_cleanup().

• "Cleanup" is clearer than "deinit", which isn't a proper word.
• The function might not be exclusively use for NN related cleanup. In some devices it could include cleanup for DSP hardware accelerator (if that exists).

regards,
Joseph

[yoursanonymous]

Thanks for the suggestion, Joseph

I have renamed th_nn_deinit() to th_cleanup() and updated all call sites accordingly. The new name is clearer and leaves room for future cleanup beyond NN-specific resources.

Best regards,
Vinayak

[joseph-yiu]

Merged to dev_2026q1 branch.
regards,
Joseph