Prevent multiple instances of shutting down the same room simultaneously

[anoadragon453]

We had a process attempt to shut down the same room using two, separate calls to the Admin API:

• DELETE /_synapse/admin/v1/rooms/{roomId}
• POST /_synapse/admin/v1/{roomId}/delete

Both of these method handlers eventually call RoomShutdownHandler.shutdown_room, and there doesn't look to be any code that deduplicates requests anywhere.

This lead to a state where two long-running shutdown processes for the same room occurred and choked the main processes' CPU, resulting in the need to restart the main process to recover its performance.

[dklimpel]

It is related to

• Room deletion (shutdown) fail in a constant loop due to non-serializable access caused by PostgreSQL isolation levels #10294
• Host dead after caches seem to break possibly due to multiple room delete commands in short period of time #10792
• Shuting down a room with 4k+ users takes more than 2h #7651

[dklimpel]

Perhaps this API should return an ID like the purge room API. https://github.com/matrix-org/synapse/blob/develop/docs/admin_api/purge_history_api.md#purge-status-query

Like this comment #7651 (comment)