Improve onion service (aka hidden service) modeling

[sporksmith]

Some things to revisit after merging #5

• Measure HS circuits per HS user: We currently calculate the number of non-HS-circuits-per-non-HS-user and then also use that as the HS-circuits-per-HS-user. Ideally we ought to more directly measure the latter, perhaps using the technique from section 6.3 of https://torusage-imc2018.github.io/paper.pdf.

• Measure fraction of HS traffic and/or users: The user currently specifies what fraction of traffic to make hidden-service traffic. Probably better to estimate the fraction of HS traffic in the real network and use that (perhaps still with a user-configurable multiplier on top).

• Generate HS markov models?: The HS traffic may be different enough in character from everything else that it'd be worth learning and using separate markov models for HS brackground traffic.

• Split HS metrics: HS metrics are currently lumped in with non-HS metrics, but are likely to be substantially different. Probably better to split these out to separate lines or graphs.

[sporksmith]

@mikeperry says:

we do have https://metrics.torproject.org/hidserv-rend-relayed-cells.html to get an idea of the amount of HS traffic compared to total network traffic (https://metrics.torproject.org/bandwidth.html), but Rob probably has better ideas

This could be helpful for the measurement items.