Unify correctness and benchmark testing with pytest

[voltjia]

Summary

• Add bench module with assert_match for correctness checking and benchmark for performance benchmarking via triton.testing.
• Split modules.py into a modules/ package with separate files for each module.
• Move max_pool2d kernel into ops/ninetoothed/ for consistency with other operators.
• Add pytest infrastructure with correctness tests (test_ops.py) and benchmark tests (test_benchmarks.py).
• Remove the 13 standalone operator scripts, since both correctness and benchmarking are now covered by pytest.

Usage

pytest # Correctness tests only.
pytest -m benchmark # Benchmark tests only.
pytest -m benchmark -k TestMM # Single operator benchmark.