fix(ios): pop only the topmost React Native screen on popToNative

[KisaneNeko]

Fixes #354.

popToNative() posts an app-wide NotificationCenter notification with no sender, so every live React Native surface acted on it — including ones sitting underneath other screens. With a stack like RN → native → RN, one call popped the lot and took the native screen in the middle with it.

There was a second, smaller problem in the same path: a SwiftUI-hosted surface handled the notification twice, once in ReactNativeView via dismiss() and again in the ReactNativeViewController it wraps via popViewController. So even a single RN screen popped one level too far.

BrownfieldSurfaceRegistry now elects one surface — the most recently shown one that's actually on screen — and only that surface acts, once per host. This matches what Android already does implicitly, since its back dispatcher only ever delivers to the top callback.

A couple of things worth knowing while reviewing:

• Visibility is read live (view.window) rather than cached from the appearance callbacks. SwiftUI recreates hosted controllers, so a cached flag is stale exactly when the notification arrives — I had it the other way round first and it broke.
• SwiftUI owns its navigation state and reverts an imperative popViewController, so the SwiftUI host still does the dismissing. The controller stands down and only steps in if the dismiss clearly did nothing, which is the case for a ReactNativeView inside a UIHostingController pushed onto a UIKit stack.

Two behaviour changes are called out in the changeset: popToNative() is now a no-op (with a log) when no RN surface is on screen, and it no longer dismisses a modally presented native screen that merely contains an RN surface.

E2E coverage and a demo screen to exercise this are in a follow-up PR stacked on this one, to keep the fix reviewable on its own.

🤖 Generated with Claude Code

[hurali97]

This looks like an unusual place to put this and related files. Let's put this at packages/react-native-brownfield/ios/.

We can then discard the Package.swift change

[KisaneNeko]

I followed the BrownfieldBundleURLResolver example. SPM can't reach outside its package root, so moving the file to ios will delete the unit tests. I can do that but just making sure we're fine with that.

For what it's worth they only cover the surface election logic, which is the simple part. The bugs I actually hit were all in the wiring and the e2e caught those, so losing them isn't a big deal.

[hurali97]

Yeah let's ignore this test.