perf: faster app startup

[bajrangCoder]

Summary

Makes the editor usable much sooner on Android and iOS, and keeps plugins, file I/O and optional features from slowing startup. App and plugin behaviour stay the same.

Results

Cold starts of a signed release on a Motorola Edge 40 Neo (Android 15) with a real set of installed plugins and developer mode on, plus the iPhone 17 Pro simulator (free edition).

	Before	After
Android: editor visible (from page start)	2,100 ms	~860 ms
Android: all plugins loaded	10.8 s	~4.0 s
Android: file opened from another app	~7.7 s	~2.3 s
iOS: editor visible	995 ms	~406 ms
main.js (Android / iOS)	3,349 / 3,356 KB	1,515 / 1,460 KB

From tap to editor on the same Android device it is now ~1.15 s, including ~430 ms for Android to show the native splash and start the WebView.

Startup sequence

• The splash hides once the app has rendered a frame instead of after a fixed 500 ms delay.
• Independent native calls (encodings, storage folders, install source, Android version) run in parallel.
• The purchase check runs alongside startup.
  • Free builds still settle Pro status before plugins initialize, and non-Pro users before the UI is usable.
  • Paid builds never wait.
  • The wait is capped at 5 s.
• Developer tools (eruda) start after the editor is visible. Console output and errors captured during startup are replayed into eruda's console, so nothing is lost. With developer mode off, the capture is dropped as soon as settings load.

Smaller startup bundle

Code that is not needed for the first frame now loads on demand:

• Terminal / xterm. It is ready before plugins load, so the synchronous plugin terminal APIs behave as before.
• Run/preview with markdown-it.
• mime-types.
• The LSP client. It loads only when an enabled server handles an open file, after the editor is visible. This covers the client manager, editor features, @codemirror/lsp-client, marked and the protocol types. The plugin-facing server registry, the diagnostics UI and the server launcher stay eager.
• Emmet. It loads, through an editor compartment, only when a file uses an Emmet syntax.

Polyfills are bundled once (native.js always loads first). swc targets the oldest supported engine instead of ES2015 (chrome >= 67 on Android, ios_saf >= 18 on iOS), with core-js in entry mode so plugins still get every global.

Less startup work

• Encodings are cached per app version and device.
• Built-in editor themes skip the load-time check; they are still validated when applied.
• Built-in sidebar app chunks load in parallel.
• Command registration no longer rebuilds the whole keymap per command. The keymap is rebuilt lazily on read and applied to editors once per burst, before the next key event.

File I/O and plugins

• UTF-8 text is decoded and encoded in the WebView instead of being sent to native and back as base64. A 3.7 MB read went from 814 ms to 9 ms with identical output. Invalid UTF-8, lone surrogates and other charsets still use native, and the BOM handling matches each platform.
• Android file URLs are formatted in JS when the result provably matches native, which saves a resolve round trip per read and per plugin path. It matched native for every path checked on device, and falls back otherwise. iOS keeps resolving natively.
• Fewer native round trips per plugin. The plugin's cache file and token are prepared while its script downloads.
• Files shared from other apps no longer wait for the slowest plugin. They open after a 1 s grace period. Files a plugin may handle (remembered handler extensions, PDF/Office documents) still wait for plugins, and order is kept.

Fixes

• A billing service that never answers can no longer keep the app on the splash screen.
• Revoked Pro status no longer returns offline.
• The owned plugins list no longer crashes (disabledMap scope).
• Cancelling the sponsor form no longer causes an unhandled rejection.