Fix reuseMaps infinite loop when reusing into the same container

[mikeyfarina]

Fixes #2613.

With reuseMaps, Mapbox.reuse() moves the pooled map's children from its old container into the mount container with while (oldContainer.childNodes.length > 0) container.appendChild(oldContainer.childNodes[0]). When the two containers are the same DOM node, the loop never terminates and the tab freezes at 100% CPU. That case is reachable whenever the <Map> mount effect re-runs while the DOM survives — React StrictMode's effect replay in development, and back/forward navigation restore (Activity semantics, e.g. Next.js 16 App Router) in production builds.

Changes, applied to all three copies of the pattern (react-mapbox, react-maplibre, mapbox-legacy):

• reuse() treats oldContainer === container as "already in place" and skips the reparenting.
• The [mapboxgl-children] cleanup moves from recycle() into reuse()'s different-container branch. Removing it at recycle time broke the same-container case a second way: reuse() returns the same instance, so setMapInstance bails out on Object.is and React never re-inserts the children div it still owns — any plain-JSX map child silently disappeared after the reuse. Removing it only when the DOM is actually moving keeps the surviving React tree intact, and the different-container behavior is unchanged (same removal, later moment).

yarn test: 52 files / 84 tests pass. The react-mapbox version of this change has also been running in production as a vendored patch since 2026-08-27, including a browser regression test that walks map pin → popup → team page → browser back (the sequence that previously froze on the first back-navigation).

https://claude.ai/code/session_01DmN5k333nk7YLAhtyHE8em

[tordans]

Do we have or should we have some kind of test file or maybe demo page that shows the error in the different cases before (and fixed after)?