TCP metrics

[simwr872]

Discussed in #110:

I'm not too familiar with eBPF but I was also wondering if the number of active TCP connections could be exposed as a metric.

The rationale being that if a metric was available, it could be forwarded by an adapter and used for scaling decisions. One use case is to enable granular scaling of bursty, latency-sensitive, workloads that cannot accurately be scaled based on cpu/memory alone.