use lapply in subsetBy

[sgibb]

Dear Laurent,

this PR replace the for loop in subsetBy (that contains a growing vector) with a lapply based solution.
This results in a slightly faster subsetting:

library("git2r")
library("devtools")

repo <- repository("MSnbase")

checkout(repo, branch="master")
load_all("MSnbase")
load_all("synapter")

load(file.path(system.file("data", package="synapterdata"), "ups25a.rda"))
ms25a <- as(ups25a, "MSnSet")

system.time({ma <- topN(ms25a, groupBy = fData(ms25a)$protein.Accession, n = 3) })
#   user  system elapsed
#  2.064   0.000   2.066

checkout(repo, branch="lapply")
load_all("MSnbase")
system.time({la <- topN(ms25a, groupBy = fData(ms25a)$protein.Accession, n = 3) })
#    user  system elapsed
#   1.200   0.008   1.207

checkout(repo, branch="split")
load_all("MSnbase")
system.time({sp <- topN(ms25a, groupBy = fData(ms25a)$protein.Accession, n = 3) })
#    user  system elapsed
#   0.692   0.004   0.695

This approach doesn't change anything (all topN tests pass) and has (hopefully) no negative side effects.

As you could see there is another branch split https://github.com/lgatto/MSnbase/blob/split/R/utils.R#L324-L338 that uses a split approach that is even faster.
The split approach returns its results in a different order (order of the levels of group) in contrast to the original one and to the lapply approach (order of appearance in group). That's why some unit tests fail and why I don't offer the split solution as PR. I think I will need your help to ensure everything is correct in the split branch.

[lgatto]

Great, thanks!
I don't see a reason why split wouldn't be adequate - there is no specification that states the order of the output. Would it be possible to explicitly set the levels to reflect their order of appearance?