feat: add weighted confidence scoring for extraction validation

[Arijit429]

Closes #60
Closes #450

🚀 Summary

This PR enhances the extraction validation workflow by introducing field-level weighted confidence scoring instead of using a flat deduction model.

The goal is to make confidence evaluation more representative of real-world data importance and improve the reliability of the requires_review decision.

---

✨ What Changed

Updated the ExtractionValidator logic to use weighted importance for each field:

FIELD_WEIGHTS = {
    "location": 30,
    "time": 20,
    "severity": 30,
    "description": 20
}

Each missing field now reduces confidence based on its relative importance instead of applying a uniform deduction.

---

💡 Why This Helps

In the previous approach, all missing fields contributed equally to confidence reduction.

This could under-represent critical missing information.

For example:

• missing location should have higher impact
• missing severity should strongly affect review decision
• missing description may be comparatively less critical

The new weighted model makes confidence scoring more realistic and production-aligned.

---

🔍 Example

Previous behavior

all fields = equal impact

New behavior

location → 30
severity → 30
time → 20
description → 20

This allows the validator to make smarter review decisions.

---

🧪 Testing

Added dedicated unit test coverage for weighted confidence scoring.

Validated scenarios include:

• multiple high-weight missing fields
• partial extraction completeness
• review threshold correctness
• confidence floor protection

Executed locally using:

PYTHONPATH=. pytest test/test_extraction_validator.py -q

and verified successful execution.

---

🎯 Impact

This improves:

• confidence score realism
• human review routing accuracy
• downstream PDF fill reliability
• production readiness of extraction workflow

[Arijit429]

Hi maintainers, this update improves the review workflow by making confidence scoring field-aware and more aligned with real extraction importance—happy to refine the weighting logic based on project needs.