We need to find at runtime which node is proxying the current
request on the primary site, so we are also saving and sending
extra JWT-signed data through headers in Workhorse, similar to
the already existing "is proxied or not" header.