Update remaining specification gaps in circuit-build-timeout spec

I should write these up while they are still somewhat fresh in my mind from my work on circuit timeouts in Arti.

• Only 3-hop circuits are considered for timeout measurements. (#48 (closed))
• Closing "idle" circuits that have never been used based on their "idle timeout".
• Closing circuits that have been dirty for too long.
• The bit about clients assuming that they have no network connectivity when they have no incoming cells is not actually done. Already done
• Breaking ties when computing most populous histograms for Xm. Already done
• Different circuit lengths and purposes get their timeouts modified based on their length -- see circuit_expire_building for the whole pile of detail.
  • Should we multiply the "abandon timeout" as well as the base timeout?
• How to reconstruct the success/failure history from the counts in the state file.
• Defaulting to abandon circuits at the default timeout * 1.5.