Bump version to 0.7.0

[JRufer]

Version bump for the 0.7.0 release, following the same shape as the 0.6.5 bump (#144).

Touches the five files that carry the version:

• Cargo.toml (workspace version)
• Cargo.lock (regenerated with cargo update --workspace, not hand-edited)
• package.json
• package-lock.json (both project-level version fields)
• src-tauri/tauri.conf.json

No code changes. It's a minor bump rather than a patch because this release adds a trigger keyword, a new setting and new mid-speech behaviour.

What 0.7.0 contains

• Add "Hey Vox" trigger keyword; tighten trigger matching #147: "Hey Vox" trigger keyword, stricter trigger matching, and no panics on non-ASCII text
• Detect voice commands while the user is still speaking #148: voice commands recognised while you're still speaking (new features.early_command_detection setting)
• Skip S1-mini cleanup for text bound for Speak targets #149: no S1-mini rewrite for text that's spoken aloud
• Start TTS model loads earlier without loading at every startup #150: TTS models start loading earlier, without loading at every startup
• Overlay: never show the target visualizer over SYSTEM RESPONDING #151: the overlay visualizer never shows on top of SYSTEM RESPONDING
• Mock wl-copy in the empty-text clipboard test #152: the routing tests no longer replace your clipboard or hang when piped

These replace 1f13d62, which was reverted on development in 29ab1f8.

Merge order: #147, then #148, then #149; the other PRs in any order; this one last.

Test plan

• Everything above merged onto development together without conflicts
• cargo test for voxctrl-app, voxctrl-config, voxctrl-routing, voxctrl-tts, voxctrl-inference: 426 pass
• npx vitest run: 244 pass
• cargo check clean, no warnings
• No 0.6.5 version strings left in the five version files
• Release build and manual smoke test of the voice-command flow

🤖 Generated with Claude Code