V2 docs update

[anikde]

🚀 Description

This PR introduces a comprehensive overhaul to the dataset documentation and parsing algorithms, ensuring alignment with the latest data schema and improving visual readability.

📝 Key Changes

1. Transition to JSON over CSV

• Deprecated the outdated .csv recognition parsing structure across the pipeline.
• Edited the dataset conversion utility script (utils/fetch_lmdb_format_data.py) to flawlessly and natively ingest the newest BSTD_v17.57.json data objects for seamless LMDB generation!

2. Task 3 (Script Identification) Integrity Fix

• Completely rewrote the Task 3 documentation block to align meticulously with the official IJDAR research paper.
• Discarded the outdated 3-way language grouping statistics table, strictly establishing it as the correct 12-way script classification task to prevent user confusion.

3. README Readability & UI Formatting Overhaul

• Vivid Data Downloads: Consolidated all dataset download links into striking, block-quoted UI blocks (> 📥 **Data Download**: ) ensuring that researchers can instantly locate datasets without deep-reading.
• Streamlined Schemas: Wrapped lengthy JSON format examples into collapsible <details> blocks to dramatically decrease visual distraction.