Fixed .shallow to consistently retain keys and indices.

[MarkusBonsch]

Closes #2336.

Keys and indices are retained correctly by shallow if retain.key = TRUE and cols != NULL.
The largest possible part of an index is retained.
If the original index was on x1, x2, x3, and `cols = c("x1", "x2"),
the shallow copy gets the same index for c("x1", "x2") (if there was a native index on x1 and x2, this is kept)

Tests have been added.

A benchmark (code at end of this post) shows that there is no negative speed impact:

master	PR
33 ms	33 ms

The improved key retainment makes it possible to clean setattr calls in foverlaps.R while still passing all tests.

Code for benchmark

library(data.table)
library(microbenchmark)

nrow <- 1e6
ncol <- 20
nindex <- 20
ncopycol <- 10

DT <- data.table(x1 = rnorm(nrow))
setindex(DT, x1)
index <- c("x1")
for(col in seq_len(ncol-1)){
  DT[, paste0("x", col + 1) := rnorm(nrow)]
  if(col < nindex){
    index <- c(index, paste0("x", col+1))
    setindexv(DT, index)
  }
}

test <- microbenchmark(shallow = data.table:::.shallow(DT, 
                                                       cols = paste0("x", seq_len(ncopycol)), 
                                                       retain.key = TRUE), 
                       times = 100, 
                       unit = "ms")

[MarkusBonsch]

Gets a problem on foverlaps test in travis that I don't get locally. Will need to investigate.

[codecov-io]

Codecov Report

Merging #2337 into master will increase coverage by 0.05%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #2337      +/-   ##
==========================================
+ Coverage   91.09%   91.14%   +0.05%     
==========================================
  Files          61       61              
  Lines       11785    11804      +19     
==========================================
+ Hits        10735    10759      +24     
+ Misses       1050     1045       -5

Impacted Files	Coverage Δ
R/foverlaps.R	94.3% <100%> (-0.11%)	⬇️
R/data.table.R	97.45% <100%> (+0.02%)	⬆️
src/rbindlist.c	88.93% <0%> (+0.19%)	⬆️
src/forder.c	94.47% <0%> (+0.52%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 3c1b6d0...52622cc. Read the comment docs.

[MarkusBonsch]

Initially, my local tests didn't include fOverlaps since genomicRanges was not installed. Now, everything should be fixed.