v0.9.2

Changed:
Added a first-run setup wizard with personal profiles, language selection, provider configuration, ChatGPT sign-in, and voice preferences.
Added secondary bindings and enable/disable controls for voice, dictation, context, screenshot, and other global hotkeys.
Added native Wayland screenshot capture and richer Linux accessibility context through AT-SPI.
Improved active-window, browser, selected-text, and document context handling.
Redesigned and expanded Settings, onboarding, overlay, bubble, and assistant workflows.
Improved local STT and Kokoro installation with pinned dependencies, progress reporting, install contracts, and Windows CUDA/cuBLAS support.
Added released-app speech diagnostics for troubleshooting local voice installations.
Added a complete self-uninstall workflow with translated UI.
Expanded Spanish, French, Simplified Chinese, and Traditional Chinese translations.
Added Azure signing for Windows release executables.
Expanded MCP documentation and clarified client/server responsibilities.
Fixed:
Fixed KDE screenshot delays, context-menu positioning, selected-text capture, and browser/application context detection.
Fixed stale or incomplete speech installations being treated as valid.
Fixed missing CUDA and cuBLAS components being detected too late.
Fixed CUDA warm-up and float16 fallback failures being silently suppressed.
Prevented STT and TTS installations from unexpectedly replacing shared dependency versions.
Fixed stale optional-package swap directories and unnecessary bundled pip files inflating releases.
Improved speech warm-up, shutdown, optional-dependency, and runtime stability.
Fixed secondary push-to-talk keys not receiving complete release-path test coverage.
Fixed release-manifest argument handling.
Fixed configuration tests leaking profile state from a developer’s local .env.
Fixed Windows-specific dependency tests failing on Linux CI.