feat: add native SQLBot-powered data extraction workflow

[2212-spc]

PR Description

Summary

This PR introduces a native structured-data extraction workflow into Open-NotebookLM and embeds the SQLBot core as an internal backend module.

The goal is to let notebook users work with structured CSV data in a multi-turn workflow:

• select notebook data sources
• ask natural-language data questions
• get SQL + result table + CSV export
• reuse generated outputs in later turns
• import generated outputs back into notebook sources

---

What’s included

1. Intelligent data extraction workflow

• Add a dedicated data_extract workflow for notebook CSV sources
• Support natural-language querying over structured data
• Return:
  • answer summary
  • generated SQL
  • tabular preview
  • CSV export

2. Multi-datasource querying

• Support selecting multiple structured data sources
• Distinguish between:
  • primary datasource
  • linked datasources
• Forward linked datasource context into the extraction flow

3. Session / turn / artifact persistence

• Persist notebook-scoped extraction sessions
• Persist per-turn messages and results
• Persist reusable output artifacts
• Support restoring historical extraction sessions

4. Artifact reuse and source import

• Generated CSV outputs can be reused as input in later turns
• Generated artifacts can be imported back into notebook sources

5. Embedded SQLBot backend

• Vendor SQLBot core under sqlbot_backend/
• Rewrite imports into isolated namespace
• Copy required SQLBot resources into the repository:
  • templates
  • few-shot examples
  • terminology data
• Add external / embedded SQLBot adapter modes
• Default extraction mode is now embedded

---

Main implementation areas

Backend

• Added notebook-scoped data_extract routes and service layer
• Added adapter layer for SQLBot execution
• Added embedded SQLBot execution path
• Added artifact/session persistence and export handling

Frontend

• Added intelligent data extraction UI to notebook workspace
• Added multi-datasource workflow interaction
• Added extraction history and reusable output artifact strip
• Added artifact import-back-to-sources flow

Embedded engine

• Added sqlbot_backend/ as a vendored structured-data engine
• Reduced import-time side effects for embedded usage
• Added configuration/adapter groundwork for in-process execution

---

Why this change

Open-NotebookLM currently works well for document-centric workflows, but structured data analysis needs a different interaction model.

This PR adds a notebook-native workflow for structured data so users can:

• query CSV data without writing SQL
• iterate across multiple turns
• reuse generated outputs as workflow inputs
• keep structured-data work inside the notebook experience

---

Current behavior / scope

This PR is focused on the data extraction workflow.

It does not attempt to expose the full SQLBot system as a general-purpose backend replacement for all Open-NotebookLM workflows.

Current scope:

• structured data extraction
• CSV-oriented multi-turn workflow
• artifact reuse/import
• embedded SQLBot path for extraction flow

---

Compatibility notes

• SQLBOT_MODE supports both:
  • external
  • embedded
• This PR defaults the extraction workflow to embedded
• Existing notebook features outside the data extraction workflow are intended to remain unaffected

---

Notes for reviewers

This is a large PR because it combines:

1. user-facing workflow additions
2. notebook-scoped persistence for extraction sessions/artifacts
3. vendoring and embedding of SQLBot core

Recommended review order:

1. fastapi_app/routers/data_extract.py
2. fastapi_app/services/data_extract_service.py
3. fastapi_app/workflow_adapters/*
4. frontend_zh/src/pages/NotebookView.tsx
5. sqlbot_backend/

---

Follow-up work

Potential follow-ups after this PR:

• continue reducing embedded-mode dependency surface
• refine SQL generation quality for count/order/top-N queries
• improve legacy artifact/history display cleanup
• split large frontend panel into smaller components