feat: comprehensive input-side incident data validation (#305)

[dhanasai2]

Pull Request: Input-Side Incident Data Validation (#305)

Overview

This PR implements comprehensive input-side validation for incident data, catching incomplete or malformed data early in the pipeline before expensive PDF generation operations.

Issue: #305

---

What's the Problem?

FireForm was missing critical validation at the input stage:

1. Transcript Input: No validation before LLM processing
2. Extracted Data: No validation before PDF generation
3. Template Configuration: No validation of field setup

This resulted in wasted resources, unclear errors, and poor user experience.

---

The Solution

A comprehensive multi-stage validation system:

TranscriptValidator

• Type checking (must be string)
• Length validation (10-50,000 chars)
• Content validation (detects incident keywords)
• Early error feedback before LLM calls

IncidentValidator

• Required field presence checking
• Empty/null/whitespace detection
• LLM not-found indicator handling ("-1")
• List value validation
• Optional format validation

Pipeline Integration

• FileManipulator.fill_form() - Validates inputs early
• Filler.fill_form() - Validates extracted data
• API routes - Validation with error handling

---

Changes Summary

File	Changes
src/validator.py	New: 400+ lines of validation logic
src/file_manipulator.py	Enhanced: Transcript & field validation
api/routes/forms.py	Enhanced: Validation in endpoint
tests/test_validator.py	New: 50+ test cases

---

Testing

• 47+ comprehensive test cases
• Type validation, required fields, empty values
• Whitespace handling, LLM "-1" indicator
• List/array validation, unicode support
• Edge cases and boundary conditions
• All tests passing

---

Usage

Simple interface:

from src.validator import validate_incident, validate_transcript

errors = validate_transcript(user_input)
if errors:
    return {"error": "Invalid input", "details": errors}

errors = validate_incident(extracted_data)  
if errors:
    raise FormValidationError(errors)

Strict interface:

result = validate_incident_strict(data)
if not result.is_valid:
    result.raise_if_invalid()  # Raises ValidationException

---

Benefits

✓ Prevent wasted API calls on invalid transcripts
✓ Catch data issues before PDF generation
✓ Clear, actionable error messages
✓ Flexible & configurable per agency
✓ Zero overhead for valid data
✓ Backward compatible

---

Fixes #305