feat: support image input in channel messages

[Umupa]

summary

Bub currently cannot process image attachments sent via Telegram or Discord. When a user sends a photo, the image is ignored and only text (if any) is passed to the model. This limits Bub's usefulness for vision-capable models.

Current behavior

• User sends a photo in Telegram/Discord
• Bub receives the update but strips the image from the message payload
• The model only sees text, or receives an empty message if no caption was provided

Expected behavior

• Bub detects image attachments in incoming channel messages
• Downloads or fetches the image URL from the channel adapter (Telegram/Discord)
• Passes the image as a image_url content block in the LLM message (following the OpenAI vision API format)
• Vision-capable models can then describe, analyze, or respond to the image