feat: add validation and review pipeline for extraction workflow

[Arijit429]

🚀 Summary

This PR strengthens FireForm’s extraction workflow by introducing a dedicated validation and review pipeline for AI-generated structured outputs. The goal of this change is to improve extraction reliability, enable safer human-in-the-loop verification, and move the system closer to a production-ready workflow.

---

✨ What Changed

Added a dedicated validator module

Created a new utility file:

src/utils/extraction_validator.py

This introduces the ExtractionValidator class responsible for:

• validating extracted incident fields
• detecting missing / malformed values
• generating confidence score
• determining whether manual review is required

---

Integrated validation into extraction workflow

Updated:

src/file_manipulator.py

The extraction flow now includes a dedicated validation stage before PDF filling.

Updated flow

Frontend Input
   ↓
Structured LLM Extraction
   ↓
Fallback to Legacy Extraction (if needed)
   ↓
Validation Layer
   ↓
Confidence Scoring + requires_review
   ↓
PDF Filling
   ↓
Final Output

---

📌 Before

Previously, extracted data was passed directly into the PDF generation workflow after structured extraction / fallback.

This could allow:

• incomplete fields
• empty values
• malformed extraction outputs
• low-confidence reports

to move forward without sufficient validation.

---

✅ After

With this PR, every extracted output now passes through a validation pipeline that:

• checks required fields
• identifies missing data
• assigns confidence score
• flags incomplete outputs for manual review

This improves both reliability and safety of the generated reports.

---

🎯 Impact

• Improves extraction consistency
• Enables human-in-the-loop verification
• Reduces incomplete report generation risk
• Strengthens production readiness
• Improves maintainability through modular validation logic

---

🧪 Testing

Tested locally using FastAPI Swagger routes.

Verified:

• successful extraction flow
• fallback extraction path
• validation output logging
• requires_review generation
• successful PDF output creation

---

🔮 Future Scope

This validation layer also creates a strong foundation for future improvements such as:

• field-level confidence scoring
• schema-based validation
• advanced NLP entity verification
• route-level validation reporting