Add Triangle Multiplicative Update problem

[josusanmartin]

Summary

• Add a new triangle-multiplicative-update problem based on GPU MODE's (Mini) Competition new problems! #3: AlphaFold's Triangle Multiplicative Update
• Define the masked outgoing TriMul contraction output[b,i,j,h] = sum_k left[b,i,k,h] * mask[b,i,k] * right[b,j,k,h] * mask[b,j,k]
• Include problem statement, PyTorch reference/baseline, test cases, verification, and FLOP accounting

Source inspiration: https://stormy-sailor-96a.notion.site/GPU-MODE-Mini-Competition-3-AlphaFold-s-Triangle-Multiplicative-Update-207221cc2ffa8034b3eddff1d898dc14

Validation

• python3 -m py_compile problems/triangle-multiplicative-update/def.py problems/triangle-multiplicative-update/torch.py
• Imported the problem class with PYTHONPATH=/home/josu/dev/tensara-selfhost/platform/engine and sanity-checked parameters, test cases, extra params, and FLOP count

Note: I did not run CUDA correctness locally because this shell does not expose a CUDA runtime.