Now, the k8s spawner spawns runtime on the same k8s cluster. We should separate them, because:

1. they serve very different purposes

   • web/api servers serves html/js/css and graphQL endpoints, and are crucial for app responsiveness
   • runtime server is stateless. It may come from different machines, e.g., our remote server, user's own server, user's browser via WASM (like JupyterLite)

2. the resource requirements are very different

   • web/api servers use small amount of resources
   • runtime servers use much more resources: e.g., 2 core 4GB instances
   • web/api servers load is more predictable in terms of number of concurrent users
   • runtime servers is much less predictable: a user could launch a large number of runtime instances