I found a lame workaround by setting up a cronjob to ping the API regularly, but there was still bad latency for cloud functions that were interacting with the database. It was 1-2s of latency, which was a total nonstarter for my realtime collaboration code. I had to go write a socket.io server by hand instead.
Also, there are extremely limited tools for observability/monitoring/logging of cloud functions, as well as for plain API interactions with firestore. I could see that trying to track down issues through that toolset was going to be nightmarish. In my opinion, for any app that scales beyond a small personal project, it is an absolute must to have good tools for inspecting logs with error rate graphs and stacktraces and so on.