Load documents, images, audio, and video using Azure Content Understanding.
Produces LangChain Document objects with extracted markdown content
and rich metadata (fields, confidence scores, source info).
Exactly one of file_path, url, or bytes_source must be provided.
