feat(rig-openai-responses): add image support for OpenAI reasoning models

[xlyk]

Summary

• Fix image handling for OpenAI reasoning models (o1, o3, o4, gpt-5)
• Images were being silently dropped when sending messages to reasoning models
• Add convert_user_content() function to properly handle multimodal content

Problem

The extract_user_text() function in rig-openai-responses only handled Text and ToolResult content types, ignoring Image content entirely. This meant images sent to reasoning models were silently dropped.

Solution

• Added new convert_user_content() function that handles all content types including images
• Converts rig's UserContent::Image to async-openai's InputImageContent format
• Uses EasyInputContent::ContentList when images are present (required by the OpenAI Responses API for multimodal content)
• Added base64 dependency for encoding raw image bytes
• Added unit tests for text-only and image content conversion

Affected Models

All OpenAI reasoning models now properly support images:

• o1, o1-preview, o1-mini
• o3, o3-mini
• o4-mini
• gpt-5, gpt-5.1, gpt-5.2, gpt-5-mini, gpt-5-nano

Test Plan

• Unit tests pass (cargo test -p rig-openai-responses)
• Clippy passes with no warnings
• E2E tests pass (119 tests)