Multi-Mode Durable(-ish) Objects

[helloimalastair]

Hey y'all!

Had some ideas for DOs(or Durable-ish Objects), and I wanted to run them by you to see whether they are good/have issues. From least to most durable:

1. Support for Hibernatable WebSocket APIs in normal Workers. This would allow you to terminate a WebSocket connection within a Worker, with a bit more safety against unexpected disconnects.

2. Storage-less DOs. Would be able to be spawned anywhere, not just where storage is able to be replicated. Could be used for coordination in cases where you don't particularly care about durable recovery(i.e. chatrooms with no history).

3. Less-Durable DOs. Similar to Storage-less DOs, in that they could be spawned in any colo. They would have access to the Transactional Storage API, but without any replication, so if that machine/instance goes down, that's it. Would be useful for cases where you handle recovery outside DOs, but you need more geographic distribution. This setup may be a bit more iffy, as it would rely on storage space being available in every colo, rather than the handful that support DOs today.