Add SQL safety validation and query intent classification to Text-to-SQL pipeline

[kriti20041]

Summary

This PR enhances the Text-to-SQL analytics pipeline by adding safety and interpretability layers on top of LLM-generated SQL queries.

Changes

• Added SQL safety validation to detect and block destructive queries (e.g., DROP, DELETE, UPDATE) before execution.
• Implemented query intent classification to distinguish aggregation queries from row-level analytics.
• Improved overall reliability and control of LLM-driven database interactions.

Motivation

LLM-generated SQL can be unsafe and difficult to interpret in production environments.
These changes introduce guardrails and reasoning layers to ensure secure, read-only execution and better understanding of user analytics intent.

Impact

• Prevents execution of destructive SQL statements
• Improves transparency of generated queries
• Makes the Text-to-SQL system more production-ready

Testing

• Manually tested with sample natural language queries
• Verified unsafe SQL queries are correctly blocked
• Confirmed intent classification for aggregation and row-level queries