Add local RAG harness: indexer, retriever, prompt injection and docs

[muzaffermahi]

Motivation

• Provide a local Retrieval-Augmented Generation (RAG) workflow so SmallCode can inject similar code snippets from indexed GitHub repos into model context without external embedding services.
• Offer an easy way to build and maintain a local snippet index and expose it to the TUI so models get concrete examples for coding/debugging tasks.

Description

• Add a lightweight local RAG index and embedding implementation in src/rag/index_store.js with tokenized hashed embeddings, cosine search, and upsert/save/load support.
• Add a retriever in src/rag/retriever.js that plans query variants, runs iterative searches, formats top hits as fenced-code blocks for prompt injection, and exposes a Google fallback URL when confidence is low.
• Add a GitHub shallow-clone scraper and chunker in src/rag/github_scraper.js and a CLI indexer bin/rag-index.js (installed as smallcode-rag-index) to build .smallcode/rag/index.json from a .smallcode/rag/repos.json repo list.
• Integrate RAG into the TUI by loading RagRetriever in bin/smallcode.js and injecting getRagContext into both the system prompt and the dynamic context path; update prompts to include RAG context when available.
• Add user-facing docs docs/rag-harness.md, README additions about quick run and indexing, package changes (package.json/package-lock.json) to expose the new bin and rag:index script, and unit tests in test/rag.test.js covering index ranking, query planning, and prompt formatting.

Testing

• Added unit tests in test/rag.test.js that exercise RagIndexStore, RagRetriever, and planQuery.
• Ran the repository test script npm test (which runs node --test test/*.test.js) and the new RAG tests passed.
• Exercised the indexer locally using the new script npm run rag:index / smallcode-rag-index to verify index creation and CLI output.

---

Codex Task