This is a high-level timeout for the http.Handler application code,
not a low-level network timeout.

Trace requests on client and server HTTP traffic, using a system-wide
configuration file to control reporting.

And return the right error code (403, not 401).

Doh, don't know how this was missed until now.

Fixes include:
* only notify systemd after the network socket has been opened
* add net/http/pprof debug handlers
* add /health endpoint for Prometheus blackbox prober

Allows users to specify a set of trusted forwarders in the
configuration, and then using github.com/gorilla/handlers.ProxyHeaders
to rewrite the http.Request parameters according to X-Forwarding-* and
X-Real-IP headers.

Reduces cyclomatic complexity score and makes code more readable.

Includes TLS authentication, to authorize peers
based on the subject of their X509 certificate.