Return a warning when g^*=NA

[osofr]

Sometimes the user may set the intervention node (g^*) to NA. If this value is actually being applied, then we should produce a warning, since the resulting weights will be also NA.

The best place to catch it is probably here:
https://github.com/osofr/stremr/blob/master/R/DeterministicBinaryOutcomeModel.R#L29-L44

    fit = function(overwrite = FALSE, data, ...) { # Move overwrite to a field? ... self$overwrite
      self$n <- data$nobs
      self$define.subset.idx(data)
      private$probA1 <- data$get.outvar(TRUE, self$gstar.Name)
      # private$.isNA.probA1 <- is.na(private$probA1)
      # self$subset_idx <- rep.int(TRUE, self$n)
      self$subset_idx <- seq_len(self$n)
      private$.outvar <- data$get.outvar(TRUE, self$getoutvarnm) # Always a vector of 0/1
      # private$.isNA.outvar <- is.na(private$.outvar)
      self$is.fitted <- TRUE
      # **********************************************************************
      # to save RAM space when doing many stacked regressions wipe out all internal data:
      # self$wipe.alldat
      # **********************************************************************
      invisible(self)
    },

Or here (private$probA1 is the g^*):
https://github.com/osofr/stremr/blob/master/R/DeterministicBinaryOutcomeModel.R#L53-L67

  predictAeqa = function(newdata, ...) { # P(A^s[i]=a^s|W^s=w^s) - calculating the likelihood for indA[i] (n vector of a`s)
      assert_that(self$is.fitted)
      if (missing(newdata)) {
        indA <- self$getoutvarval
      } else {
        indA <- newdata$get.outvar(self$getsubset, self$getoutvarnm) # Always a vector of 0/1
      }
      assert_that(is.integerish(indA)) # check that observed exposure is always a vector of integers
      probAeqa <- rep.int(1L, self$n) # for missing values, the likelihood is always set to P(A = a) = 1.
      # probA1 <- private$probA1[self$getsubset]
      probA1 <- private$probA1
      probAeqa[self$getsubset] <- probA1^(indA) * (1 - probA1)^(1L - indA)
      self$wipe.alldat # to save RAM space when doing many stacked regressions wipe out all internal data:
      return(probAeqa)
    },
`