[FEAT]: Add input-side incident data validation gate

[geoffkats]

📝 Description

Add an input-side validation layer to reject incomplete incident data before PDF generation. This validates that required fields like incident_type, location, and time are present and non-empty before data proceeds further in the pipeline.

💡 Rationale

Currently, incomplete or malformed incident data can enter the pipeline and lead to invalid PDF generation or downstream failures. Adding a validation gate improves reliability and prevents bad data from reaching the PDF filler stage.

🛠️ Proposed Solution

• Logic change in src/
• Update to requirements.txt
• New prompt for Mistral/Ollama

Implementation outline:

• Add validate_incident() in src/validator.py
• Validate required fields before PDF filling
• Return deterministic error messages for missing or empty fields
• Add unit tests for valid input, missing fields, empty fields, whitespace-only fields, and invalid input type
• Integrate validation into src/file_manipulator.py

✅ Acceptance Criteria

• Feature works in Docker container.
• Documentation updated in docs/.
• Input incident data is validated before PDF fill.

Definition of done:

• validate_incident() returns [] for valid incident data
• Missing required fields produce deterministic error messages
• Empty or whitespace-only required fields produce deterministic error messages
• Unit tests cover all validation cases and pass locally
• Validation runs before PDF fill logic

📌 Additional Context

PR already opened: #304

This is distinct from #114:

• This contribution validates input completeness before PDF filling
• [FEAT]: Schema Validation + Error Recovery #114 focuses on output/schema validation after LLM extraction