Add file content part support for multimodal chat completions

[adambalogh]

Summary

Adds support for file content parts in multimodal chat completion requests, enabling the gateway to handle PDF attachments and other file inputs alongside text and image content. This is a prerequisite for the "secure attachments" feature where PDF files arrive as OpenAI file content parts.

Changes

• OpenAPI Schema: Added ChatCompletionRequestMessageContentPartFile schema with nested file object supporting three input modes:

  • file_data: Base64-encoded inline file data (with optional filename)
  • file_id: Reference to a previously uploaded file
  • filename: Name hint for inline data

  Updated the ChatCompletionRequestUserMessageContentPart oneOf union to include the new file part type alongside existing text, image, and audio parts.

• Test Coverage: Added TestUserMessageContentPartValidation unit test class with schema-validation regression tests:

  • test_image_part_passes_schema_validation: Verifies image content parts pass validation (baseline)
  • test_pdf_file_part_passes_schema_validation: Verifies file content parts with PDF data pass validation

  Tests use a standalone _build_app() helper to build a connexion app directly from the OpenAPI spec, avoiding the optional flask_testing dependency and ensuring validation runs in the lean test environment.

Implementation Details

• File parts are validated at the connexion request-body validation layer, the same pipeline that processes OHTTP inner requests
• The schema allows flexible file input modes (inline base64, file ID, or filename) to support various attachment workflows
• Tests confirm that PDFs no longer receive 400 validation errors (they may fail later with 500 if provider keys are missing, but that's expected in unit tests)

https://claude.ai/code/session_01HF83iVseKjA673AM4n4uN2