feat(backend): add confidence score for each extracted field

[Godatcode]

Summary

Closes #60

• Add a confidence score (0.0–1.0) to each LLM-extracted field, changing the output format from raw values to {"value": "...", "confidence": 0.92}
• Update the Ollama prompt and payload ("format": "json") to request structured JSON responses
• Add parse_llm_response() with clamping and graceful fallback for malformed responses
• Log [WARNING] for fields with confidence < 0.5
• Ensure Fill.fill_form() still passes plain strings to PDF filling

Changes

• src/backend.py: Updated build_prompt(), main_loop(), add_response_to_json(); added parse_llm_response() and get_confidence_report()
• src/test/test_confidence.py: 17 unit tests covering JSON parsing, confidence clamping, fallback behavior, storage format, and warning output

Test plan

• All 17 unit tests pass: PYTHONPATH=src python -m pytest src/test/test_confidence.py -v
• End-to-end test with Ollama: verify [LOG] Resulting JSON shows {"value": ..., "confidence": ...} format
• Verify [WARNING] appears for low-confidence fields
• Verify filled PDF contains correct plain string values (not stringified dicts)

[Godatcode]

@juanalvv @vharkins @marcvergees — requesting your review on this PR. It adds confidence scores to the LLM extraction pipeline (issue #60). Happy to address any feedback!

[Godatcode]

@juanalvv @vharkins1 @marcvergees Could you review this when you get a chance? Happy to address any feedback.