Enable data-related args to survival::coxph()

[hfrick]

Several arguments to survival::coxph() capture variables or variable names which are to be found in the data: id, cluster, istate. Those don't work with censored/parsnip yet:

library(censored)
#> Loading required package: parsnip
#> Loading required package: survival

# `id` arg
mod <- coxph(Surv(tstart, tstop, status) ~ age + steroids + strata(sex),
             data = cgd, id = id)

spec <- proportional_hazards() %>%
  set_engine("survival", id = id)
f_fit <- spec %>%
  fit(Surv(tstart, tstop, status) ~ age + steroids + strata(sex),
      data = cgd)
#> Error in model.frame.default(formula = Surv(tstart, tstop, status) ~ age + : invalid type (language) for variable '(id)'

# `cluster` arg
bladder1 <- bladder[bladder$enum < 5, ] 
mod <- coxph(Surv(stop, event) ~ (rx + size + number) * strata(enum),
             cluster = id, bladder1)

# but also possible to provide `cluster` as a vector
mod <- coxph(Surv(stop, event) ~ (rx + size + number) * strata(enum),
             cluster = bladder1$id, bladder1)

spec <- proportional_hazards() %>%
  set_engine("survival", cluster = id)
f_fit <- spec %>%
  fit(Surv(stop, event) ~ (rx + size + number) * strata(enum),
      data = bladder1)
#> Error in model.frame.default(formula = Surv(stop, event) ~ (rx + size + : invalid type (language) for variable '(cluster)'

^{Created on 2022-02-02 by the reprex package (v2.0.1)}

[hfrick]

For prediction / survival curves: the id column will need to be present in new_data, similar to how the stratification variable needs to be present.

[hfrick]

check id_var() in multilevelmod

[hfrick]

tidymodels/parsnip#713 would fix the issue with fitting the model but would still need to solve the issue at prediction time

[hfrick]

Also tidymodels/stacks#117 for dealing with parsnip's quosures since tidymodels/parsnip#737 put an end to tidymodels/parsnip#713 mentioned above