feat(computer-use): integrate cua-driver-rs v0.6.8 for enhanced background automation

[bobleer]

Summary

Integrates key source code from cua-driver-rs v0.6.8 to enhance BitFun's built-in Computer Use capability. This adopts the open-source community's strongest Computer Use techniques for both macOS and Windows background automation.

What's Changed

macOS Background Input (fully wired)

• SkyLight SPI bridge (macos_skylight.rs): Loads private macOS framework symbols via dlopen/dlsym for background input (SLEventPostToPid, activate_without_raise, etc.)
• Dual-post strategy (macos_bg_input.rs): Events posted via BOTH SkyLight SLEventPostToPid AND public CGEvent::post_to_pid for maximum compatibility
• Chromium click recipe (bg_click_chromium): 5-event sequence with routing field stamping (f0 phase, f1 clickState, f40 pid, f51/f91/f92 windowID, f58 click-group) — solves the long-standing Chromium/Electron apps ignoring background clicks issue
• Focus-without-raise (activate_without_raise SPI): Uses real CGWindowID from CGWindowListCopyWindowInfo to focus target app without raising its window
• Terminal-safe typing: Detects terminal emulators by bundle ID, routes to per-keystroke key-event synthesis
• AX tree improvements (macos_ax_dump.rs): AXChildren ∪ AXWindows union for background app tree completeness; Chromium AX enablement via AXManualAccessibility/AXEnhancedUserInterface; AXPlaceholderValue fallback
• AX focus (macos_ax_write.rs): try_ax_focus wired into app_type_text for reliable text input

Windows Background Input & UI Automation (fully wired)

• UIA batched cache (windows_ax_ui.rs): IUIAutomationCacheRequest with TreeScope_Subtree + ControlViewCondition() for efficient tree walking
• Windows AX tree snapshot: get_app_state_snapshot with foreground pid, MSAA fallback for SAL/VCL windows
• Background input (windows_bg_input.rs): post_click_screen, post_scroll_screen, post_drag_screen, inject_text_cloaked, inject_key_cloaked with VK parsing
• Window capture (windows_capture.rs): PrintWindow/BitBlt with DWM crop; WindowCapture geometry for PointerMap construction
• App enumeration (windows_list_apps.rs): EnumWindows + process image name resolution
• Desktop host wiring (desktop_host.rs): app_click, app_type_text, app_scroll, app_key_chord, app_wait_for, interactive/visual views, list_apps, foreground window screenshot attach
• Cross-platform drag: ComputerUseHost::drag trait method with macOS bg_drag and Windows post_drag_screen overrides

Cross-Platform Infrastructure

• Element token system (element_token.rs): Per-pid LRU snapshot registry with opaque s{hex}:{idx} tokens — wired into get_app_state_inner on both platforms
• Debug overlay (debug_overlay.rs): annotate_screenshot_with_click for visual click verification in debug builds
• Chromium/Electron detection (is_chromium_electron): Bundle ID matching against Chrome, Chromium, Electron, Brave, Edge, Arc, etc.

Code Quality

• All new code formatted with cargo fmt / pnpm run fmt:rs
• cargo check -p bitfun-desktop passes clean on Windows
• 87 computer_use unit tests + 20 element_token tests pass (macOS CI baseline)
• No breaking changes to existing API surface

Test Plan

• cargo check -p bitfun-desktop passes (Windows verified)
• cargo test — 87 computer_use + 20 element_token tests pass
• Code formatted with pnpm run fmt:rs
• Manual macOS testing: background click on Chromium/Electron apps
• Manual macOS testing: background text input in Terminal/iTerm
• Manual Windows testing: get_app_state, app_click, app_type_text, drag on foreground app

Source Attribution

Integrated techniques and algorithms from cua-driver-rs v0.6.8 by the trycua team, licensed under MIT.