Make sanity checks in `marker-detection.Rmd` more robust

[PeteHaitch]

From #bioc-builds > Books now being built on macOS in devel? @ 💬

stopifnot(which("CD3E"==rownames(ordered)) > 10000), stopifnot(all(c("GNLY", "NKG7") %in% head(rownames(ordered)))), stopifnot(all(c("LYZ", "S100A9", "S100A8") %in% head(rownames(ordered)))) and stopifnot(out["LYZ", "median.AUC"] > 0.9) in OSCA.basic/marker-detection.Rmd fail on kjohnson3 (I don't know about lconway). Do these errors suggest that the results on kjohnson3 are qualitatively different from what we get on Linux? This would not be good.

These sanity checks have been manually tweaked over the years to accommodate slight changes in the marker gene results.
Need to make these checks more robust.