[Feature] Fix startup overlay: voice toggle plus autocomplete and screen intelligence debug

[senamakel]

Summary

Make the openhuman-overlay experience reliable when the app starts, and use the floating panel to expose a clear voice on/off control alongside debug visibility for predictive autocomplete and screen intelligence so developers and power users can verify behavior without digging through logs alone.

Problem

The overlay is launched best-effort from the core when the RPC server starts (src/openhuman/overlay/process.rs), but startup behavior and the panel contents are not meeting the bar: users cannot depend on voice being easy to toggle from the overlay, and debugging autocomplete vs screen intelligence is awkward or incomplete in the UI. That blocks validation of voice flows and screen-intelligence work in real desktop sessions.

Solution (optional)

• Startup / lifecycle: Ensure the overlay launches consistently in supported dev and packaged layouts (binary discovery, ordering relative to core, failure surfacing). Adjust Tauri overlay UI as needed.
• Voice: Add or fix a visible toggle that turns voice capture / dictation pipeline on and off, with state that matches the actual audio path.
• Debug: In the same overlay surface, expose actionable debug for autocomplete (e.g. context/suggestions path) and screen intelligence (feature status, last events, errors) without requiring separate tools—scoped so normal users are not overwhelmed (e.g. dev-only section or compact indicators).

Coordinate with screen intelligence / autocomplete RPC and macOS-only constraints documented in the engine (src/openhuman/screen_intelligence/).

Acceptance criteria

• Reliable at launch — On app start in supported configurations, the overlay appears when expected; if it cannot start, behavior is logged and/or user-visible per product rules.
• Voice toggle — User can see and use an on/off (or equivalent) control for voice from the overlay; state reflects reality (not purely cosmetic).
• Combined debug — Autocomplete and screen intelligence each have debug affordances in the overlay (or clearly linked from it) suitable for diagnosing integration issues during development and support.

Related

• Overlay spawn: src/openhuman/overlay/process.rs
• Screen intelligence / autocomplete: src/openhuman/screen_intelligence/
• Umbrella voice work (if tracked): [Feature] Integrate Open Whisper-style voice dictation and commands into the desktop UI #332 and sub-issues