Infinity Loop v1.1.0

Added

• Multiple AI providers. In addition to Google Gemini, the app now supports
  OpenAI (ChatGPT), Anthropic (Claude), xAI (Grok), NVIDIA NIM, and any local
  OpenAI-compatible server via Ollama. The active provider and model are chosen
  in Settings.
• Custom model entry. The model field accepts any model id, not just the ones
  in the suggestion list.
• Custom window title bar with drag-to-move, double-click to maximize, and
  minimize / maximize / close controls. Native window decorations are disabled.
• Cross-platform release builds. A tagged push (v*) triggers a GitHub Actions
  workflow that produces installers for Windows (NSIS), macOS (dmg, Apple
  Silicon and Intel), and Linux (AppImage, deb), and attaches them to a draft
  GitHub Release whose body is filled from the matching section of this file.
• Continuous integration workflow running svelte-check, cargo fmt,
  cargo clippy, and cargo test on every push and pull request.
• Rust unit tests covering provider resolution, request-body construction, and
  streaming delta parsing.

Changed

• All API keys are now stored and used exclusively in the Rust backend and are
  persisted to a config.json file in the OS application-config directory. The
  frontend only ever learns whether a key is set, never its value.
• All LLM network calls (main search, Dev-mode helpers, and PDF analysis) now
  run in the Rust backend rather than the web view. Streaming is delivered to
  the frontend over a Tauri channel.
• Request cancellation is handled by the backend via a request id instead of an
  AbortController in the web view.

Security

• API keys are no longer stored in localStorage or sent from the frontend.
  Keys entered in earlier versions are migrated into the backend on first launch
  and removed from localStorage.