[TimeToBuildBob]

Summary

Fixes #16432. The Linux/macOS shortcut script keeps its counter in the document, while native zoom survives reload. After 140% → reload, zoom-in incorrectly requests 120%.

• Track zoom per Webview, shared across cloned handles and document recreation. Absolute Rust/JS setters and relative shortcuts use the same state; shortcut limits remain 0.2–10 and absolute setters remain unclamped.
• Send relative changes through the existing set_webview_zoom command/ACL; existing numeric payloads are unchanged. Handle the command synchronously so rapid reset/shortcut operations keep arrival order.
• Apply zoom state mutations on the main thread, where Wry applies native setters. Locking around enqueueing alone lets queued worker setters diverge from newer main-thread setters.
• Resolve main-thread Wry zoom by stable webview ID, avoiding the parent-ID mutex that worker-thread reparenting can hold while waiting for the event loop.
• Add four Rust state/order regressions, four dependency-free Node script tests (Linux/macOS × keyboard/wheel), a CI step, and a changeset.

Verification

• cargo test -p tauri --lib --no-default-features --features test --offline: 84 passed on Linux.
• node --test crates/tauri/src/webview/scripts/zoom-hotkey.test.cjs: 4 passed. All four fail with the original script.
• cargo clippy -p tauri -p tauri-runtime-wry --lib --no-default-features --features tauri/test --offline: passes with one pre-existing redundant-clone warning in manager/webview.rs:272.
• Scoped rustfmt, Prettier and git diff --check pass.
• Standalone native Linux/WebKitGTK fixture built against this checkout: 800 → 666 → 571 CSS pixels; real document reload retains 571; next zoom-in reaches 500 (160%); reset returns 800. Existing released Tauri 2.12.5 reverses to 666 at that step.
• Native fixture also passes absolute numeric JS setter → shortcut, Rust set_zoom → shortcut, five-event burst, reset, and a queued-worker/main-thread interleaving. The wheel-handler test uses a synthetic DOM wheel event in the native webview, not physical wheel input.
• Restoring the pre-main-thread shadow update makes the new execution-order Rust regression fail (3.0 versus expected 2.0). Restoring Wry's parent-ID-locking setter makes the native concurrent reparent/zoom fixture fail its return marker; the repaired version returns without deadlock.

Boundaries

Native evidence is Linux only. macOS is script-test coverage; Windows/macOS native and mobile cross-compilation remain CI/maintainer verification. This does not persist zoom across process restarts or track direct platform mutations via with_webview, outside Tauri's setters. No ActivityWatch-specific polyfill or reload reset is added.

The commit is signed. I read the contribution policy and will not generate replies to review comments; any requested review discussion needs a human handoff.