Work around dplyr 0.8 handling of factors in groups

[billdenney]

Related to tidyverse/dplyr#4061

This causes a failure for PKNCA tests at least with tss.monoexponential at

pknca/tests/testthat/test-time.to.steady.state.R

Lines 296 to 324 in d2cd503

	expect_equal(
	expect_warning(
	pk.tss.monoexponential(
	conc=tmpdata$conc,
	time=tmpdata$time,
	subject=tmpdata$subject,
	treatment=tmpdata$treatment,
	time.dosing=0:14,
	verbose=FALSE
	)
	),
	data.frame(subject=factor(c(1, 10, 2:9)),
	tss.monoexponential.population=4.57618156812974,
	tss.monoexponential.popind=c(
	5.14156352865421, 4.64862524830397, 4.45956707917941,
	4.41492203844343, 4.6782583033301, 4.0823047621517,
	4.96242115751172, 4.52424147509819, 3.70338406668837,
	5.1465280219363),
	treatment=c("A", "B", "A", "A", "A",
	"A", "B", "B", "B", "B"),
	tss.monoexponential.individual=c(
	5.87784329336254, 4.71066285661623, 4.51882509145954,
	3.91269286106442, 4.74475071729459, 3.99341726779716,
	5.08737230904342, 4.50068650719192, 3.4876172020751,
	5.35051537086801),
	tss.monoexponential.single=4.56067603534),
	tolerance=1e-4,
	check.attributes=FALSE,
	info="pk.tss.monoexponential 1")

It also probably causes issues with the summary functions, but testing currently doesn't include factors with imbalanced levels.

[billdenney]

This fix will likely affect #74 and #79, too.

[billdenney]

As discussed in the dplyr issue, the behavior is reverted to the prior behavior so no change is required.