fix: dependency tidy

[goodroot]

Closes #178

[github-actions]

Continue PR Review

Automated review via Continue CLI run through opub.

Summary

The change improves error handling around text injection and adds targeted tests, but the new GNOME/Wayland direct-typing fallback has a functional regression risk for non-ASCII and layout-dependent text input, and the new tests do not cover that behavior.

Findings

• On GNOME/Wayland, _inject_via_clipboard_and_hotkey() now prefers ydotool type before any clipboard-based paste path (lib/src/text_injector.py). This bypasses the clipboard path entirely for those sessions, but ydotool type is layout-dependent and typically unreliable for Unicode/non-ASCII text, dead keys, and IME/composed input. Users dictating accented characters, symbols, or non-Latin scripts may get corrupted output where clipboard paste previously preserved exact text. This needs either a capability check, a narrower fallback, or tests that prove correct behavior for non-ASCII input.
• The new direct-typing path also changes semantics around “paste” vs “type” for terminal/editor targets on GNOME/Wayland. Clipboard paste preserved multiline text and exact bytes, while ydotool type replays keystrokes, which can trigger editor/terminal keybindings, auto-pairing, shell behaviors, or different handling of tabs/newlines. There are no tests covering multiline input, tabs, or special characters through the GNOME fallback, so this is a regression risk that should be covered before merge.

Suggested follow-ups

• Add tests for GNOME/Wayland direct typing with Unicode/non-ASCII text, multiline text, tabs, and punctuation-heavy input.
• Consider restricting the ydotool type fallback to plain ASCII/single-line text, and keep clipboard-based behavior for anything else.
• Add an integration-style test or documented compatibility matrix for common GNOME targets (GTK text fields, terminals, browsers, Electron apps).

[goodroot]

The above has some nuance which makes me happy with the implementation as is. Merging.