Right now, if any of our HTTP handlers hits a panic, we silently swallow it with no logs. Not great for detecting errors!

Also the HTTP client calling the endpoint just gets no response, and it would be nice to reply with a 500 Internal Server Error.