evaluation pipeline implemented

[FacuSentena]

PR: SQL Translation Evaluation Pipeline

🚀 Summary

Implemented an automated framework to benchmark Snowflake → Databricks SQL translation models using MLflow and LLM-as-a-judge.

✨ Key Features

• Strict Scoring: Deduction-based scoring for Compliance (functional correctness) and Best Practices (optimization/docs).
• A/B Comparison: Uses Nested MLflow Runs to enable side-by-side grouped bar charts for model comparison.
• Diagnostics: Exports top_issues_summary.txt and issues_table.json to pinpoint model weaknesses.
• Interfaces:
  • CLI: run_local_benchmark.py for batch runs.
  • Notebook: benchmark_interactive.ipynb for visual research.

📁 Changes

• Added src/artifact_translation_package/evaluation/ module.
• Consolidated documentation in evaluation README.md.
• Updated requirements.txt with MLflow and Databricks integrations.