Include a sample worker config for a Big deployment

[ara4n]

e.g. Matrix.org or Mozilla or KDE, so that folks can crib off how to set up redis & workers & haproxy with real data rather than having to guess it from workers.md. See https://twitter.com/uumas8/status/1316469781932003333 for more context.

[callahad]

We think it could be useful to base this on matrix.org as a maximal example, as we split out everything we can there.

[behrmann]

I recently hacked together a proof of concept for the mapping of users to the always same backend worker, which is currently left as an exercise to the reader.

The rest of the config is guess work from how I read the load balancing section in the worker docs and I haven't tested the federation part, since I only run non-federated instance, but I'd be happy to know if this is how it's supposed to work. If that were the case I could polish this up as a PR.

[behrmann]

The new token format will make load balancing a lot easier in the long term, since the local part of the mxid is a easier to extract (it would have been even easier, if the base64 part would have been padded), but is there a way to age out old tokens? Currently one would need to handle both old and new style tokens.