fix(plugins): hide the window before running quit hooks (#94)

[kipavy]

Fixes #94. Based on @Flash303's diagnosis and patch in #95 — the approach here is theirs, with two extra safety fixes.

Cause

ensureQuitHandler() (src/plugins/runtime.ts) intercepts onCloseRequested, calls preventDefault(), and then awaits every plugin onBeforeQuit callback (capped at 5s) before triggering the Rust-side exit. The window stays on screen for that entire wait. The gist-sync plugin pushes to GitHub in its quit hook, so closing the app looked frozen for ~2s. With no plugin registering a quit hook the handler is never installed at all, which is why this only reproduces with a plugin like gist-sync enabled.

Change

• Hide the window before awaiting the hooks, so the wait is invisible and the app appears to close instantly. Requires the core:window:allow-hide capability (not part of core:default).
• Only hide when a quit hook is actually registered; with none, close immediately as before.
• Arm the Rust-side exit up front with a fallback timer and call it from a finally block. Without this, a throwing hook or an invoke that never lands would leave a hidden window over a live process — worse than a slow close.
• Wrap hook invocation in an async function so a synchronous throw becomes a rejection instead of escaping the handler.

Tests

New src/plugins/runtime.quitHandler.test.ts covers: hide precedes the hook, no hide when there are zero hooks, exit still happens when a hook throws, and exit on the 5s cap with the fallback timer disarmed (force_quit invoked exactly once). pnpm tsc --noEmit clean; pnpm vitest run src/plugins green (73 files, 593 tests).

Live verification

Debug build driven under Xvfb/WebKitGTK through tauri-driver, with a test plugin whose quit hook runs for ~2.4s:

• Close fired, hook ran, plugin:window|is_visible sampled from inside the hook: false at every sample from 303ms onward through 2429ms — the window is gone while the sync is still running.
• Process exited 2.97s after the close was fired, i.e. right after the hook finished, not on the 6s fallback.
• With the hook unregistered, the process exited 240ms after the close was fired (that figure includes the harness round-trip), and no hide occurred.

Not verified on Windows — no Windows machine here. The mechanism is platform-independent, but @Flash303 confirmed the original symptom and fix on Windows 11.

[Flash303]

Not verified on Windows — no Windows machine here. The mechanism is platform-independent, but @Flash303 confirmed the original symptom and fix on Windows 11.

Works well on Windows, thanks and waiting for the release !!