Drop HTTP tracing layer as we do not really use it.

Since we do not yet use distributed tracing, all this yields is an endless stream of

ERROR tower_http::trace::on_failure: response failed classification=Status code: 500 Internal Server Error latency=1 ms

log messages which we filter out anyway.

I think if we do see elevated errors levels, we should rather use the logging facilities of Caddy to figure things out.