Fix Responses API image forwarding

[City-Zero]

Problem

Requests to POST /v1/responses that include an input_image part are accepted, but the image is not forwarded in the upstream structured contents field. Multimodal models therefore receive only the accompanying text.

Fix

• Preserve input_image parts while converting Responses user messages.
• Convert them to upstream image_url content parts.
• Keep the existing text conversion unchanged.

Verification

• Added a regression test for a Responses message containing both input_text and input_image.
• go test ./...
• go vet ./...
• Manually verified /v1/responses with a base64 PNG using dfmodel.

Anthropic Messages and OpenAI Chat multimodal paths were also checked and already forward image content correctly.

[Liki4]

Problem

Requests to POST /v1/responses that include an input_image part are accepted, but the image is not forwarded in the upstream structured contents field. Multimodal models therefore receive only the accompanying text.

Fix

• Preserve input_image parts while converting Responses user messages.
• Convert them to upstream image_url content parts.
• Keep the existing text conversion unchanged.

Verification

• Added a regression test for a Responses message containing both input_text and input_image.
• go test ./...
• go vet ./...
• Manually verified /v1/responses with a base64 PNG using dfmodel.

Anthropic Messages and OpenAI Chat multimodal paths were also checked and already forward image content correctly.

oh jesus 我没意料到会有人用 orz