fix(models): include multimodal chat models in discovery

[xxx7xxxx]

Summary

• Include multimodal conversational tasks in model discovery so available vision, audio, video, and omni models appear in the chat catalog.
• Parse comma-separated task values inside arrays and preserve availability filtering and vision metadata.
• Verified gateway discovery, catalog persistence, and reload behavior locally.

[xxx7xxxx]

Close as the AI Gateway will fix the glitch.