We should log the start and stop of all requests, especially the long-lived ones like /watch (which we call for the current workspace before connecting to a remote) and /watch-metadata (which we call for every running agent on every user-owned workspace in the sidebar, although this should only happen while the sidebar is displaying).

So probably a request ID, endpoint, method, and payload for the start log, and then a request ID, status code, response body, and duration for the end log.

Also I think our logs are not timestamped right now? sorted, we use the built-in logger now

This should help us narrow down reports of leaks, since it looks like these connections may not be disposing when they should.