Rewrite OcrService to match the actual inference API response format
(nested status/result structure with PascalCase/snake_case field names).
Register IOcrService in DI with typed HttpClient and Polly v8 resilience
(retry, timeout, circuit breaker via AddStandardResilienceHandler).
Key changes:
- Fix response model to match real API (InferenceApiResponse)
- Map correct field names (InvoiceNumber, InvoiceDueDate, OCR, Amount, etc.)
- Add extract_line_items=true for VAT summary extraction
- Copy stream before sending to avoid disposal conflicts with retries
- Add JsonException handling for malformed responses
- Remove sensitive data from error logs
- Add 35 unit tests covering field mapping, VAT parsing, error handling,
decimal/date formats, and content type detection
