add support for a list of selection expressions in .SDcols

[r2evans]

Closes #6619.

This adds the ability to select columns with more than one type of expression.

BLUF:

DT = data.table(int1=1L, int2=2L, chr1="A", chr2="B", num1=1, num2=2, lgl=TRUE)
DT[, .SD, .SDcols=.(is.numeric, patterns("2$"), "lgl")]
#     int1  int2  num1  num2   chr2    lgl
#    <int> <int> <num> <num> <char> <lgcl>
# 1:     1     2     1     2      B   TRUE

Justification

Traditional column-selection allows selection by:

• column names
• integers (column index)
• patterns(..)
• function call, e.g., is.numeric

In order to combine two or more of these in traditional methods, we would need to first run the functions or patterns individually and then either cbind them all or do similar efforts to collect column names and pass those into a single .SDcols=cols argument.

If we want to select all numeric columns, those ending in "2", and the "lgl" column, then we would need to do one of these methods:

cbind(DT[, .SD, .SDcols=is.numeric], DT[, .SD, .SDcols=patterns("2$")], DT[, .SD, .SDcols="lgl"])
#     int1  int2  num1  num2  int2   chr2  num2    lgl
#    <int> <int> <num> <num> <int> <char> <num> <lgcl>
# 1:     1     2     1     2     2      B     2   TRUE

This naive approach results in duplicate selections in int2 and num2.

Another approach might be:

cols <- unique(c(names(which(sapply(DT, is.numeric))), grep("2$", names(DT), value=TRUE), "lgl"))
DT[, .SD, .SDcols=cols]
#     int1  int2  num1  num2   chr2    lgl
#    <int> <int> <num> <num> <char> <lgcl>
# 1:     1     2     1     2      B   TRUE

While the latter is not egregious, my patch enables an inline all-at-once selection that seems intuitive