this commit swaps out the clustering logic

Instead of server side clustering that's triggered on each bounding box change we download all available
features (for both offices and events) and let mapbox-gl cluster it.

drawback: on high feature count initial load is slower, as the whole payload needs to be downloaded upfront (ca. 350kb / 1000 items)
improvement: vastly enhanced snappiness of clustering, also improved representation, although that is somewhat unrelated
ref CAC_FRONTEND-325