[unitaryhack] Support arbitrary input sizes in parallel QPIXL decoding

[rashidraihan]

Fixes issue #2

Issue:
The pipeline crashed or distorted images when inputs were sized differently, as the Walsh–Hadamard transform takes power‑of‑2, same vector lengths.

Fix:

• Pad every image individually to the nearest square power‑of‑2 (pad2d_to_pow2).
• Flatten in column-major (Fortran) order and reshape again in the same order.
• Reshape after decoding to the padded square, and then crop to the original size.

Followed on 33×40 & 41×34 images; decoded results now exactly like originals with no errors or stripes.

[danielbultrini]

Thanks for your submission rashidraihan! The fix works, but sadly someone submitted an equally working fix earlier than you. I hope that our decision is understandable.

[rashidraihan]

Thanks a lot for the update! While I’m a bit sad I wasn’t first, I'm grateful for the opportunity to work on the issue and make my very first contribution. I'm glad my fix was valid. I'll definitely keep an eye out for other opportunities to contribute in the future.