fix: update DocIntel default and surface OCR failures

[imadreamerboy]

This PR fixes Azure Document Intelligence handling for image OCR in markitdown.

There were two separate problems:

1. DocumentIntelligenceConverter still defaulted to api_version="2024-07-31-preview", which can fail on valid Azure resources with 404 Resource not found during begin_analyze_document(...).
2. That failure could be masked by fallback behavior in MarkItDown._convert(): after the DocIntel converter failed, ImageConverter could return an empty DocumentConverterResult(markdown=""), and markitdown treated that as a successful conversion. The caller then saw result.text_content == "" instead of the real Azure error.

Changes

• Updated the default Azure Document Intelligence API version from 2024-07-31-preview to 2024-11-30
• Kept explicit docintel_api_version=... override behavior intact
• Changed conversion flow so an empty fallback result does not count as success if an earlier converter already failed
• Added regression tests for:
  • new default DocIntel API version
  • explicit API version override
  • empty image fallback no longer masking a prior converter failure

Why

This matches current Azure behavior more reliably for OCR/image analysis and fixes a misleading failure mode where real Azure/DocIntel errors were swallowed and surfaced as “no OCR text extracted”.

Validation

Tested with focused pytest coverage for DocIntel and fallback behavior:

• test_docintel_default_api_version
• test_docintel_explicit_api_version
• test_empty_image_fallback_does_not_mask_prior_failure

These pass with the local package under test.

[imadreamerboy]

@microsoft-github-policy-service agree