Support sending images to LLMs (vision/multimodal input)

[jrswab]

Problem

Axe currently only supports plain text content in messages to LLMs. The Message.Content field is a string, so there's no way to send images for vision/multimodal tasks (e.g., asking an LLM to describe an image or extract data from a screenshot).

Scope

1. Extend Message.Content from a plain string to a type that can hold both text blocks and image blocks (matching Anthropic content[] / OpenAI content[] formats).
2. Update provider implementations (Anthropic, OpenAI, Ollama, Bedrock, OpenCode) to serialize image content blocks in their respective API formats.
3. Add an input mechanism — e.g., a --image flag or supporting image file paths in agent config, with binary detection and base64 encoding in the resolve layer.

Notes

• Not all models support vision — graceful error handling needed when a model doesn't support image input.
• Consider supported formats (PNG, JPEG, GIF, WebP) and size limits per provider.
• Sub-agent delegation should be able to pass image context if needed.