Clustering based on tile distance

Hello! I was wondering if it's possible today to create clusters based on anything else other than the pixel distance? Mapbox has a good example where they use [`clusterRadius`](https://docs.mapbox.com/mapbox-gl-js/style-spec/#sources-geojson-clusterRadius). The difference is that clustering based on radius remains consistent irrespective of the projection and zoom level. As a result, clusters are more predictable and nicer looking. When I use the 3d projection in Cesium, clusters shift uncontrollably as I rotate the globe. This is not a desired behaviour since what was previously a cluster of 10 might become a cluster of 6, then 13 as I slowly spin the globe. Here is a [video](https://docs.mapbox.com/ios/assets/ClusteringExample-753eb7c22ae34785178d492001a63398.mp4) of the desired final effect. If this is not a current feature and not on the roadmap, would you be open to a pull request?

Start by locating Cesium's current clustering entry point and reviewing how its pixel-distance behavior is defined. Compare the requested tile- or radius-based semantics with the linked Mapbox clusterRadius example; done means clusters remain stable as projection and zoom change, including while rotating the globe.

Written by the indexing model from the issue text.