Use data.table for grouped sort in pw_info()

[jdblischak]

The profiling indicated that dplyr::group_by() and dplyr::arrange() were two of the biggest bottlenecks:

I confirmed these were the ones in pw_info():

gsDesign2/R/pw_info.R

Lines 207 to 210 in 6d4371a

	ans <- ans %>%
	group_by(time, stratum) %>%
	arrange(t, .by_group = TRUE) %>%
	ungroup()

In my previous PR #295, I had delayed converting this to data.table. Once I remembered that group_by() also orders the columns, I was able to figure it out. Here is example code to show that the two approaches are identical:

set.seed(1)
d <- data.frame(
  x = sample(letters[1:3], size = 100, replace = TRUE),
  y = sample(letters[4:5], size = 100, replace = TRUE),
  z = sample(1:5, size = 100, replace = TRUE)
)

library("dplyr")
x1 <- d %>% group_by(x, y) %>% arrange(z, .by_group = TRUE) %>% ungroup()
x1 <- as.data.frame(x1)

library("data.table")
x2 <- as.data.table(d)
setorder(x2, x, y)
x2 <- x2[order(z), .SD, by = .(x, y)]
setDF(x2)

all.equal(x1, x2)

In repeated benchmarking of gs_power_ahr(), this update removes reduces the runtime by 70-140 ms

I also removed an isolated dplyr::transmute() from gs_info_rd() and updated the documentation to reflect that now a data frame is returned

[jdblischak]

xref: #219

[keaven]

Thanks for the update