Revamp Extractly Streamlit UX and modular pipeline

[Yureehh]

Motivation

• Make the existing Extractly proof-of-concept feel like a client-ready product with a modern landing page and clear IA (Schema → Upload → Results).
• Preserve the core schema → classification → extraction happy path while improving reliability, traceability, and developer ergonomics.
• Introduce a modular architecture to separate UI, domain/pipeline logic, and integrations for easier maintenance and extension.
• Add sensible defaults, config management, logging, and lightweight persistence so demos and tests are reproducible.

Description

• Rebuilt the Streamlit front-end with a product-like landing page and navigation pages: Home.py, pages/1_Schema_Studio.py, pages/2_Extract.py, pages/3_Results.py, and pages/4_Settings.py.
• Added a modular backend under src/extractly/ including config.py, logging.py, domain/ (models, validation, schema_store.py, run_store.py), integrations/ (OpenAI client wrapper, OCR, preprocess), pipeline/ (classification, extraction, runner), and ui/components.py.
• Kept and adapted existing extraction/classifier wrappers (src/extractor.py, src/classifier.py) to call the new pipeline, added sample artifacts in data/sample_docs/ and schemas/demo_invoice.json, and added .env.example and README updates.
• Added basic automated tests for schema validation and run persistence in tests/ and a RunStore abstraction that persists runs to ./runs.

Testing

• Ran unit tests with pytest -q which executed the new test suite.
• All automated tests passed: 3 passed.
• No other automated checks were executed in this run.

---

Codex Task