Summary

• Replace legacy /api/inject-key flow with syncKeysToProviders() — API keys saved in the browser are now pushed to all matching server-side providers via the /api/providers REST endpoints, with async save feedback (spinner + success/error)
• Update curated model list — default changed to Kimi K2.5; added MiniMax M2.5, GLM 5, Qwen 3.5 397B, GPT-OSS 120B; removed GPT-OSS 20B, Nemotron 3 Super, Qwen3 Next 80B, Llama 3.3 70B; most now route through nvidia-endpoints
• Sync Quick Selects into model selector — custom Quick Select shortcuts from the Inference tab now appear in the model dropdown immediately on add/remove
• Enable streaming by default — patch streaming: true for the active model on bootstrap and on every model switch
• Fix tooltip clipping — left-align tooltips and change policy card overflow from hidden to visible
• Minor UI tweaks — rename "Network Policies" → "Allowed Network Policies"; update API key link to build.nvidia.com/settings/api-keys; use placeholder "-" for onboard model ID