codegen: add MatMul operator support

[strimo378]

Motivation

• Provide end-to-end support for the ONNX MatMul operator including lowering, C code emission, and runtime execution.
• Generate shape-aware C kernels for matrix multiply to keep emitted signatures precise and deterministic.
• Ensure correctness by adding unit/golden tests and ORT comparison for MatMul.

Description

• Add a MatMulOp dataclass and extend LoweredModel with input_shapes so codegen can emit shape-specific signatures.
• Implement _lower_matmul in compiler.py with 2D shape validation and inner-dimension checks, and add _apply_matmul for runtime execution.
• Extend CEmitter to render MatMulOp using a new templates/matmul_op.c.j2 template and include MatMulOp in op resolution and wrapper emission.
• Add golden output tests/golden/matmul_model.c and tests in tests/test_add_mvp.py to validate emitted code and numerical equality against ONNX Runtime.

Testing

• Ran pytest -n auto -q and all tests passed (10 passed).
• New tests include a golden file comparison (matmul_model.c) and an ONNX Runtime numeric comparison for MatMul.

---

Codex Task