feat(rag): adopt LlamaIndex for document ingestion and semantic retrieval

[keli-wen]

Summary

• Keep deterministic PDF parsing in quantmind.preprocess.format and move opinionated document chunking/retrieval into the dedicated quantmind.rag package.
• Use LlamaIndex SentenceSplitter and BM25 for page-aware document evidence, and LlamaIndex for the private LocalKnowledgeLibrary semantic-ranking backend, while preserving QuantMind public types and SQLite canonical storage.
• Enforce the new package boundary with import-linter, synchronize the canonical design under contexts/design/rag/, and move the focused example/tests to examples/rag/ and tests/rag/.

Design Boundary

quantmind.rag is an opinionated LlamaIndex data-plane package, not a generic RAG framework. It may depend on quantmind.preprocess; preprocessing and quantmind.library must not depend on it. LlamaIndex types remain private, and the package does not add public retriever, vector-store, provider, backend, query-engine, or plugin abstractions. A future PageIndex adapter may live beside these document operations without being forced through LocalKnowledgeLibrary.search().

Related Issue

Closes #117.

Verification

• ./scripts/verify.sh — 330 passed, 86.56% coverage; Ruff, basedpyright, and all 7 import contracts passed.
• uv run python scripts/verify_pdf_rag_e2e.py — passed with 15 pages, 33 chunks, and page 5 as the top result.
• uv run python examples/rag/paper.py — passed.
• Pre-commit and pre-push hooks passed without bypasses.

Checklist

• The title uses English Conventional Commit format: type(scope): summary.
• The related issue or design discussion is linked when applicable.
• ./scripts/verify.sh passes.
• Every applicable live-network component smoke test passes, or this PR states why none applies.
• Public behavior has focused tests, an example, and documentation where applicable.
• The PR is complete, small, and contains no unrelated changes.