Moonshot provider: images dropped for kimi-k3 despite native vision support

Goose refuses to send image attachments when the active provider is Moonshot with kimi-k3, replying [image omitted: model does not support vision]. Kimi K3 has native vision: Moonshot's model card states it understands text, images and video in a single model, and OpenRouter lists it as multimodal. The capability metadata for this provider/model appears to be incorrect.

References:

https://huggingface.co/moonshotai/Kimi-K3
https://openrouter.ai/moonshotai/kimi-k3

**To Reproduce**

1. Configure the Moonshot provider with model kimi-k3 and set it as active provider
2. Start a new session and attach any image
3. The image is dropped before the request is sent, with the message [image omitted: model does not support vision]

Moonshot exposes only one model id (kimi-k3), so there is no separate vision variant to select instead.

**Expected behavior**

The image should be sent to the Moonshot API, as it is when using the Alibaba provider with qwen3.8-max.

**Please provide the following information**

**OS & Arch:** macOS, Apple Silicon (M4 Max)
**Interface:** UI
**Version:** 1.48.0
**Extensions enabled:** Analyze, Apps, Developer, Extension Manager, Summon, Scheduler, Top Of Mind, Chat Recall, Todo, Skills, Memory
**Provider & Model:** Moonshot – kimi-k3

**Additional context**

Switching the active provider to Alibaba with qwen3.8-max works as expected, so the issue looks specific to the Moonshot provider's capability declaration rather than to image handling in general.