TODO

# short term
1. Finish PopulationTree reference class, add plotting functions, tests
2. Bootstrap precision matrix
3. Refactor main fitting function

# longer term
1. Streamlined rate calculation without using edge diffs
2. Add active parameter detection: mask parameters where
  - demographic parameters are Inf (coalescence) or 0 (migration rates)
  - admixture coefficient vectors are unitary
  - be sure that penalty calculation works here -- may need to set parts of gradient to 0 by hand
3. Prior on population size at time 0
4. Godambe information matrix?
5. Hessian --> Laplace approximation
6. Easier cross validation
  - automate splitting into test/train in rate calculation
  - automate refitting model across grid of penalty values