Tweak connection timeouts for Prometheus in Usage Ping

https://app.periscopedata.com/app/gitlab/679200/Topology-Dashboard?widget=9253421&udv=0 suggests that we infrequently (0.5% of all cases so far) spend ~1m in sending topology data from Usage Ping. These outliers are most likely connection or network timeout related.

We should review our timeout settings for PrometheusClient and Gitlab::HTTP which it uses under the hood. They should be very strict (I'd say 5-10 seconds) since otherwise we will block all of Usage Ping and it's unlikely for such requests to ever succeed.