add patsy formulas for Surv() syntax

[jburos]

This is a revised PR for the add-patsy-formulas feature (PR #60).

Since then:

1. Improved code formatting
2. fix log_t_dur spec (Possible log_t_dur misspecification in Stan model file #71)
3. fix dataframe merge error on pp_surv (Dataframe Merge error in utils.plot_pp_survival (Test pem_survival_model with simulated data notebook) #76)

Comments from previous PR:

Some code refactoring in order to support using patsy formulas for LHS as well as RHS spec. Closes #52

Models can now use the following syntax:

survivalstan.fit_stan_survival_model(
    formula = 'surv(event_status=end_failure, time=t) ~ age + sex',
   data = d,
   model = survivalstan.models.exp_survival_model,
    ...
)

Optional parameters to surv include group= and subject=.

The previous syntax is still also supported:

survivalstan.fit_stan_survival_model(
   event_col = 'end_failure', 
   time_col = 't',
   formula = ' ~ age + sex',
   ...
) 

Moved a fair amount of data-processing into the survivalstan.formulas module, which extends patsy syntax to support the above. For example, the patsy.ModelDesc module includes the stan-data dict which gets passed to pystan.stan for sampling.

This ipynb tests early versions of this code and may help demonstrate how the elements in formulas.py fit together.

[coveralls]

Coverage increased (+41.9%) to 85.081% when pulling d3f8c1a on v2.0 into c1e1417 on dev.

[coveralls]

Coverage increased (+41.9%) to 85.081% when pulling d3f8c1a on v2.0 into c1e1417 on dev.

[coveralls]

Coverage increased (+41.9%) to 85.081% when pulling d3f8c1a on v2.0 into c1e1417 on dev.