add graph randomization / sybil resistance cost function term discussion

[nothingmuch]

https://gist.github.com/nothingmuch/f5b9a559958c6116606d9da0d4d884f2

got some LLM suggestions for two simpler estimators that i still need to evaluate, but they look to be a better balance than all of the ones i suggested and empirically perform pretty well as well

linking the gist for now so that this is recorded somewhere before i do the work of updating and simplifying it