fix(linux): prefer dotool for reliable Wayland text input

[aasmall]

Before Submitting This PR

• I have searched existing issues and pull requests (including closed ones) to ensure this isn't a duplicate
• I have read CONTRIBUTING.md

Human Written Description

wtype works unreliably on wayland, often dropping characters. dotool was much more reliable. this PR adds it as a selectable option as well as hardens its implementation.

Prior art and context

dotool support on Linux was already discussed and accepted. This PR builds on that foundation:

• Discussion #363 — "Supporting paste action on Wayland with dotool" is where the approach was originally proposed. @cjpais's response: "Would be happy to have a PR regarding dottool."
• Discussion #742 — "Enable the ability to detect the current available typing methods" was a follow-up from a Niri user where wtype didn't work and dotool did. This motivated adding a user-facing selector.
• PR #760 (merged by @kakapt) added the TypingTool enum and the Advanced > Experimental > Typing Tool dropdown. This PR builds directly on that infrastructure — the dropdown already exists; this PR makes the dotool path much more reliable when selected.

Related Issues/Discussions

Relevant to several open bugs around unreliable text input on Linux/Wayland:

• #429 — First character missing on GNOME/Wayland Direct paste. The issue author specifically suggests dotool as a solution.
• #315 — Typing fails due to window focus timing. dotool bypasses the virtual-keyboard protocol, sidestepping focus-related character loss.
• #439 — Direct paste uses wrong keyboard layout. dotool may also address this since it operates at the uinput level.

Changes

• Prefer dotoolc (daemon client, zero startup cost) with graceful fallback to standalone dotool when dotoold isn't running
• Sanitize transcript text — strip all control characters to prevent dotool command injection via newlines
• Capture stderr from the child process and surface it in error messages (keeps diagnostics that bare status codes would lose)
• Poll-based timeout with child process cleanup on expiry (no orphaned threads or zombies). Timeout scales with text length and configured delay so long transcriptions aren't cut off mid-type; floored at max(delay_ms, 1) so delay_ms=0 still gets per-char headroom. Final try_wait() before declaring timeout prevents false positives when the child exits right at the deadline.
• New typing_delay_ms setting (0–50ms, default 2ms) with UI slider in Advanced > Experimental, with translations across all 20 locales
• Log warning on invalid typing_delay_ms values instead of silently falling back
• Show actionable warning when no native typing tools are detected (previously a quiet info log)

Auto-detection priority unchanged. On Wayland the order remains wtype → dotool → ydotool, matching upstream behavior. dotool was already in the priority list via PR #760 — this PR just makes it more capable when selected or when the auto path lands on it.

Files changed: clipboard.rs (main logic), settings.rs (new setting + tests), shortcut/mod.rs (command registration), lib.rs (command wiring), bindings.ts, settingsStore.ts, TypingDelay.tsx (new slider), AdvancedSettings.tsx, translation.json + 19 locale files

This PR is rebased directly on upstream/main and does not depend on any other open PR.

Testing

• Tested on Framework 13 AMD (Ryzen AI 300, Radeon 880M) running Arch Linux with Hyprland
• dotool typing: fast, accurate output with zero dropped characters
• Typing delay slider: verified 0ms through 50ms range, default 2ms is smooth
• Fallback: tested with dotool present, with only wtype present, and with neither — warning displays correctly
• Long transcription test: 2000+ chars at 50ms delay completes without timeout

All auto-triggered CI checks verified locally before marking ready:

• cargo fmt --check, cargo test (9 clipboard tests pass, including the new timeout scaling test)
• bun run lint, bun run format:check (excluding upstream's pre-existing AGENTS.md prettier issue)
• bun run check:translations — all 20 languages have complete keys
• bun run test:playwright — passes

Screenshots/Videos

AI Assistance

• AI was used (please describe below)

If AI was used:

• Tools used: Claude Code (Claude Opus 4.7), with code review from CodeRabbit, OpenAI Codex (GPT-5.2), and Grok
• How extensively: Claude Opus wrote the implementation with human direction on architecture (dotool as opt-in enhancement rather than priority reorder, configurable typing_delay_ms, sanitization approach, dynamic timeout scaling). Multiple rounds of independent AI code review — every actionable finding was either addressed or consciously declined with a documented reason. Translations for the 19 non-English locales generated by Claude. Human tested all changes on real hardware and made scope decisions.