Add TinyTeach: PDF-based curiosity-driven tutor app

[Ad-gh-26]

What is TinyTeach?

TinyTeach is a standalone Python/Streamlit app that turns any programming-language
PDF (Python, Java, C++, Rust, Go, etc.) into a personal tutor. It parses and indexes
the book locally with FAISS, then uses a small LLM to generate 5–8 progressive case
studies plus a learning roadmap for any topic the user asks about.

Key features

• Accepts any programming PDF — language behavior is handled through a pluggable
  prompt template registry, so the retrieval core stays language-agnostic
• Generates 5–8 case studies per topic, ordered novice → intermediate → advanced
• Each case study includes a cross-domain real-world analogy, starter code,
  expected output, progressive hints, and a fact check
• Produces a learning roadmap with milestones mapped to case studies
• Supports 4 pluggable free LLM providers: HF Inference Router, HF Local, Groq,
  and Ollama
• Grounded generation — if a topic isn't covered in the uploaded book, it says so
  instead of making something up
• Case studies and roadmap can be exported as JSON or Markdown
• No infrastructure required — runs on a free HuggingFace Space

Why it's added as a separate folder

TinyTeach runs on a different stack (Python/Streamlit) than the existing
client//server/ app in this repo, so it's added as a self-contained project
under /tinyteach rather than merged into the existing structure.

What's included

• Full source code (src/) — parsing, chunking, embedding, retrieval, generation,
  and UI layers
• Test suite (tests/) — unit and integration tests
• Documentation (docs/, README.md, PROJECT_BLUEPRINT.md, context.md)
• .env.example for configuration (no real secrets included)

What's excluded

• .env (contains local secrets)
• data/ (FAISS indices — regenerated automatically at runtime from uploaded PDFs)
• Cache/log files (__pycache__, .pytest_cache, *.log)