Move away from socket.io
According to !1 (comment 163985199), it doesn't seem that the current Go ports (both server an client) of the Socket.io library, are reliable enough to build this service.
We need to figure out what to do next. We can implement the service using bare websockets, which will imply creating a custom message protocol between the FE and the server, or we can put an extra effort and switch to gRPC (as described in #2 (closed)).
In the end, it seems gRPC would be the best option but, since we want a first iteration of the service working soon, is it worth it to reimplement the current websockets approach?
/cc @nick.thomas @DouweM