Add document management and runtime visibility

[1816x]

Summary

• add Next.js proxy routes for live documents and runtime statistics
• upload Markdown or plain text from the browser, or paste content directly
• list indexed documents and refresh corpus state after every successful insert
• surface document/chunk totals, embedding dimension and backend, cosine metric, HNSW M and ef_construction, and the active relevance floor
• distinguish lexical hashed retrieval from semantic embeddings
• distinguish extractive mock answers from Claude generation
• improve empty, loading, success, and error states across the workspace
• document the in-memory behavior and extend service regression coverage

Runtime visibility

The UI presents two independent modes:

• retrieval — lexical/hashed or semantic embeddings
• answer generation — extractive mock or Claude

HNSW indexing and search remain real in every mode.

Test plan

GitHub Actions should run:

• cargo fmt --all --check
• cargo clippy --all-targets -- -D warnings
• cargo test
• bindings pytest
• service pytest, including the expanded /stats contract
• Next.js production build and TypeScript validation

Scope

This PR intentionally does not add upload authorization, request-size limits, configurable CORS, or Claude timeouts. Those remain isolated in Tanda 3.