Add parallelism to speed up filter_by_quadkey

[dshkol]

This PR adds the option for users to use parallelism via the built-in parallel::mcmapply function as a drop-in replacement for mapply in filter_by_quadkey. This significantly speeds up running filter_by_quadkey:

oot <- get_performance_tiles(service = "mobile", year = 2020, quarter = 1, col_select = c("quadkey", "avg_d_kbps"))
nc <- sf::st_read(system.file("gpkg/nc.gpkg", package="sf"), quiet = TRUE)
snc <- nc[1,] # a reduced geo object for quicker tests

ncoot <- filter_by_quadkey(oot, bbox = sf::st_bbox(snc))
ncoot_par <- filter_by_quadkey_par(oot, bbox = sf::st_bbox(snc), parallel = TRUE)

> identical(ncoot, ncoot_par)
[1] TRUE

par_mb <- microbenchmark(
  filter_by_quadkey(oot, bbox = sf::st_bbox(snc)),
  filter_by_quadkey_par(oot, bbox = sf::st_bbox(snc), parallel = TRUE),
  times = 100
)

> par_mb
Unit: seconds
                                                                 expr      min       lq    mean   median       uq      max neval
                      filter_by_quadkey(oot, bbox = sf::st_bbox(snc)) 5.690945 6.617803 6.88801 6.880184 7.212670 7.915984   100
 filter_by_quadkey_par(oot, bbox = sf::st_bbox(snc), parallel = TRUE) 1.967439 2.589206 3.22361 3.018073 3.635814 7.065481   100

The functionfilter_by_quadkey takes the same tiles and bbox args and adds two more: parallel = FALSE and ncores = NULL). Parallelism is turned off by default because R's built in parallel often causes issues on Windows machines. A check of the operating system is triggered and returns a warning on Windows machines.

The original function is revised with a conditional flow depending on whether parallelism is enabled. When set to TRUE, the original logic is replaced like this:

if(!parallel) {
    quadkeys <- mapply(tileXYToQuadKey, xTile = tile_grid$tiles$x, yTile = tile_grid$tiles$y, MoreArgs = list(z = 16L))
  } else {
    quadkeys <- parallel::mcmapply(tileXYToQuadKey, xTile = tile_grid$tiles$x, yTile = tile_grid$tiles$y, MoreArgs = list(z = 16L),
                         SIMPLIFY = TRUE, USE.NAMES = TRUE,
                         mc.preschedule = TRUE, mc.set.seed = TRUE,
                         mc.silent = FALSE, mc.cores = ncores, mc.cleanup = TRUE)

If parallelism is selected, the number of active threads is set to max available minus one using parallel::detectCores(). This can be overriden using the ncores argument in the function.

[PedroAQCoutinho]

Very useful !
Thanks