fix(windows): use CoWaitForMultipleHandles for ARM64 WebView2 deadlock

[npiesco]

Summary

Replaces mpsc::channel + wait_with_pump pattern with CoWaitForMultipleHandles in three WebView2 call sites, fixing deterministic deadlock on Windows ARM64 and latent reentrancy issue on all architectures.

Closes #1665. Related: #583.

Problem

On Windows ARM64 (Snapdragon X Elite, aarch64-pc-windows-msvc), creating second WebView2 controller from main STA thread deadlocks inside MsgWaitForMultipleObjectsEx. Root cause is that webview2_com::wait_with_pump runs nested Win32 message pump (GetMessage/PeekMessage), which does not allow COM to re-enter apartment to deliver async completion callbacks. On x86/x64 this usually works by accident because WebView2's internal RPC piggybacks on window messages, but on ARM64 timing exposes bug deterministically.

Root cause behind #583 (deadlock creating new window from an IPC handler), which was worked around but never fixed at source.

Solution

Replace all three mpsc::channel + wait_with_pump call sites with:

ust CoWaitForMultipleHandles( COWAIT_DISPATCH_CALLS | COWAIT_DISPATCH_WINDOW_MESSAGES, INFINITE, &[event], )

COM-sanctioned mechanism for yielding STA thread while preserving reentrancy:

• COWAIT_DISPATCH_CALLS dispatches incoming COM calls (required for WebView2 async completion callbacks)
• COWAIT_DISPATCH_WINDOW_MESSAGES dispatches window messages (keeps the event loop responsive)

Approach is consistent with Microsoft's WebView2 threading model documentation on reentrancy.

Design notes

• Intentional departure from mpsc::channel: previous pattern used mpsc::channel + wait_with_pump (and comments explained why wait_for_async was avoided). PR removes approach entirely because underlying Win32 message pump prevents COM STA reentrancy, exact mechanism WebView2 needs to deliver completion callbacks.
• Rc<RefCell<_>> instead of Arc<Mutex<_>>: completion callbacks run on same STA thread (dispatched by CoWaitForMultipleHandles), so Rc/RefCell is correct COM types inside aren't Send/Sync.
• Win32_Security feature gate: Required by windows crate's CreateEventW signature (SECURITY_ATTRIBUTES). Verified won't compile without it.
• Fixes all architectures: ARM64 just exposed bug deterministically; reentrancy problem exists on x86/x64 too (see Deadlock creating a new window on the IPC handler on Windows #583).

Affected call sites

1. create_environment() CreateCoreWebView2EnvironmentWithOptions
2. create_controller() CreateCoreWebView2Controller / CreateCoreWebView2ControllerWithOptions
3. cookies_inner() CookieManager().GetCookies

Testing

• 6 unit tests added covering co_wait_for_handle behavior, the Rc<RefCell<Option<T>>> + event signaling pattern, STA reentrancy dispatch, and full create_environment/create_controller integration
• cargo test passes (8 passed, 0 failed)
• cargo clippy clean (0 warnings)
• Tested on Windows ARM64 (Snapdragon X Elite) deadlock resolved
• Tested on WSL2/Linux zero regressions (webview2 module is #[cfg(target_os = windows)])

Minimal Reproduction

https://github.com/npiesco/wry-arm64-deadlock

• main branch: deadlocks on ARM64 (uses upstream wry)
• fix/vendored-wry-tauri branch: works (uses patched wry fork)