predict() should not require response variable to be present in newdata

[danielinteractive]

Reprex:

library("mmrm")

fit <- mmrm(
  formula = FEV1 ~ RACE + SEX + ARMCD * AVISIT + us(AVISIT | USUBJID),
  data = fev_data
)

predict(fit, subset(fev_data, select = - FEV1))

fails with:

Error in eval(predvars, data, env) : object 'FEV1' not found

because h_mmrm_tmb_data() is invoked and this uses the full_formula which includes the response variable from the fit.

Reported originally in vincentarelbundock/marginaleffects#1000 (comment)

[clarkliming]

as by default we are conducting the "conditional prediction", not the "unconditional prediction", so the response is always required by default.

[danielinteractive]

Thanks @clarkliming , I still think that this is a bug to the user. It should in that case just default to "unconditional prediction"?