Factor to indicators error message

[tq21]

Fix issue #285. In addition, I added a check to ensure all covariates and outcomes have more than one category (categorical variable) or unique value (continuous variable). Code will fail immediately if users try to define an sl3_Task with data that contains columns with one value. An error message indicating the names of those columns will be printed.

[rachaelvp]

For covariates and outcome(s) that are constant (i.e., have one unique value/level), we only need to throw a warning. The error here is not necessary, e.g. imagine if someone creates a task for prediction with just one observation

[rachaelvp]

Hi @nhejazi and/or @jeremyrcoyle could you please take a look at the test errors here? I am not sure what's going on. Perhaps an error with GitHub Actions R with Java, or maybe we need to change the Java parameters for Lrnr_bartMachine? I don't think the changes in this PR are causing the error.

[nhejazi]

@rachaelvp and @tq21 sorry I missed this (not sure how), but the failure seems only be related to a weird issue with installing devtools (maybe the package was somehow briefly unavailable on CRAN, as the same issue affects builds for #398). I'm not sure why devtools is even present under Suggests -- it seems to only appear in blocks like this (

sl3/tests/testthat/test-gam.R

Lines 5 to 15 in 2bc71d8

	if (FALSE) {
	setwd("..")
	setwd("..")
	getwd()
	library("devtools")
	document()
	load_all("./") # load all R files in /R and datasets in /data. Ignores NAMESPACE:
	devtools::check() # runs full check
	setwd("..")
	install("sl3", build_vignettes = FALSE, dependencies = FALSE) # INSTALL W/ devtools:
	}

) that precede some units tests and that really shouldn't be included at all

[nhejazi]

@rachaelvp if you've reviewed these changes and they look good, then my suggestion would be to merge into devel and then fix on the backend, as per my comment for #398; happy to help