OpenAI proxy multimodal parity

[chindris-mihai-alexandru]

Problem: OpenAI-compatible proxy should fully support image and PDF inputs with correct content parts and validation.

Tasks:

• Validate content parts for image_url and file (puter_path) in /v1/chat/completions.
• Ensure PDFs pass through to Puter driver with expected schema.
• Add regression tests for image+pdf requests (stream and non-stream).

Acceptance:

• Image and PDF requests work end-to-end via proxy.
• Clear errors for invalid content parts.