Move SmolLM2 analysis files into smollm2 subfolder

Add trained hybrid extractor model (64% accuracy)

Add operator-aware splitting for word number extraction

Switch HybridExtractor from 256-way softmax to digit prediction

Add HybridExtractor for digit lookup + word number learning

Add PositionalDigitExtractor with direct digit supervision

Add SmolLM2-360M architecture analysis, fix PositionExtractor tokenization

Refactor LLM integration: consolidate training scripts, add extraction strategies

Remove verbose batch progress output from train_llm.py

Add LLM hidden state extraction training (train_llm.py)

Clarify proof-of-concept status: circuit validation complete, LLM integration in progress

Validate proof of concept: 100% arithmetic fitness with frozen circuits

Add LLM integration proof-of-concept framework and baseline evaluation