feat(reasoning): add ML model conversion pattern detection

Adds comprehensive ML model conversion pattern detector with support for:
- Framework detection (PyTorch, TensorFlow, ONNX, TensorRT, CoreML, TFLite, etc.)
- Quantization level detection (FP32, FP16, BF16, INT8, INT4, mixed)
- Built-in patterns for common conversions (PyTorch→ONNX, ONNX→TensorRT, etc.)
- HuggingFace Transformers to ONNX pattern
- Detailed guidance with code examples, pitfalls, and validation steps

All 12 tests passing.

Author: njfio