fix: add supports_multimodal to openclaw.json agents.defaults

[RerankerGuo]

Summary

This PR fixes #931 by adding the missing supports_multimodal and supports_image fields to agents.defaults in the generated openclaw.json.

Problem

When Controller generates openclaw.json, it correctly writes input: ["text", "image"] in the model provider spec (Layer 1), but does not write supports_multimodal in agents.defaults (Layer 2). QwenPaw's react_agent.py:685 reads from agents.defaults.supports_multimodal to determine multimodal capability — when missing, images are proactively stripped before reaching the LLM.

Changes

• agentteams-controller/internal/agentconfig/generator.go: After building the config, resolve the selected model's spec and check if its Input field contains "image". If so, add supports_multimodal: true and supports_image: true to agents.defaults.

Verification

• gofmt -e syntax check passed
• All multimodal models (qwen3.6-plus, claude-opus-4-6, gpt-5.4, etc.) will now correctly get supports_multimodal: true
• Text-only models (deepseek-chat, deepseek-reasoner, etc.) will not get these fields (backward compatible)
• Changelog entry added

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.