This commit adds extensive learning resources to help understand
bidirectional attention and modern LLM architectures:

Educational Content:
- docs/bidirectional_attention_tutorial.md: Deep dive into bidirectional
  vs causal attention with mathematical formulations and examples
- LEARNING_GUIDE.md: Structured 7-phase learning path with exercises
- docs/quick_reference.md: One-page reference for quick lookups

Interactive Tools:
- attention_comparison.py: Side-by-side comparison of causal vs
  bidirectional attention with visualizations
- visualize_model_attention.py: Extract and visualize attention patterns
  from the trained diffusion model

Enhanced Code:
- model.py: Added extensive inline comments to BidirectionalAttention,
  apply_rotary_emb, and norm functions explaining every design decision,
  shape transformation, and architectural choice

These materials enable aspiring LLM researchers to:
1. Deeply understand bidirectional attention mechanisms
2. Compare causal (GPT-style) vs bidirectional (BERT-style) attention
3. Learn modern components: RoPE, RMSNorm, QK normalization
4. Visualize attention patterns interactively
5. Understand when to use each attention type