chore: release v0.17.1

[release-plz-mold]

🤖 New release

• mold-ai-core: 0.17.0 -> 0.17.1
• mold-ai-db: 0.17.0 -> 0.17.1
• mold-ai-catalog: 0.17.0 -> 0.17.1
• mold-ai-inference: 0.17.0 -> 0.17.1
• mold-ai-server: 0.17.0 -> 0.17.1
• mold-ai-discord: 0.17.0 -> 0.17.1
• mold-ai-tui: 0.17.0 -> 0.17.1
• mold-ai: 0.17.0 -> 0.17.1

Changelog

Fixed

• Desktop post-generation upscaling now auto-downloads on multi-GPU hosts. A first-use Real-ESRGAN selection is pulled and configured by the remote server before its queue dispatches the generation to a GPU worker, matching the existing single-worker and standalone-upscale behavior instead of briefly failing with “upscaler model … is not downloaded.”
• Desktop: model names stay fully readable in the Generate picker. Long installed-model names now wrap instead of ending in an ellipsis, with multi-host availability moved to a secondary line so it no longer crowds out the model identity. The selected model control also grows when needed rather than clipping its value.
• Desktop generation sizes now recognize model-native recommendations. Qwen-Image and Qwen-Image-Edit expose their full recommended bucket list, including native 1328×1328, instead of calling those defaults Custom. The Size control also shows a live aspect-ratio/orientation diagram for preset and custom dimensions, and the empty Generate canvas now presents a clear first-print placeholder instead of an ambiguous black surface.

---

This PR was generated with release-plz.