faster numeric and POSIXct methods for IDate and ITime

[jangorecki]

Closes #1392.
PR not ready as consistency of timezone behaviour was not tested. I've added timezone consistency to current ITime behaviour but this limits the speed-up to UTC timezone of POSIXct type.
More tests welcome, Ideally according to Daylight saving time and time zone best practices.
Below current benchmark.

• *.default are the original methods
• the generics will dispatch to my new methods

library(data.table)
library(microbenchmarkCore) # install.packages("microbenchmarkCore", repos="https://olafmersmann.github.io/drat")
as.ITime.default = data.table:::as.ITime.default
as.IDate.default = data.table:::as.IDate.default
intpx = function(x) as.integer(as.POSIXct(x, origin = "1970-01-01", tz = "UTC"))
set.seed(1)
ii = sample(seq(intpx("2012-10-12"), intpx("2015-10-12")), 1e7, TRUE)
nn = as.numeric(ii)
ct = as.POSIXct(nn, origin = "1970-01-01", tz = "UTC")

microbenchmark(times = 10L,
               as.IDate.default(ii, origin="1970-01-01"),
               as.IDate.default(nn, origin="1970-01-01"),
               as.IDate.default(ct),
               as.IDate(ii),
               as.IDate(nn),
               as.IDate(ct))
# Unit: milliseconds
#                                         expr      min       lq     mean   median       uq      max neval
#  as.IDate.default(ii, origin = "1970-01-01") 122.3949 123.5557 136.8920 125.4423 150.1868 173.9983    10
#  as.IDate.default(nn, origin = "1970-01-01") 116.5835 117.9173 123.4366 118.3104 121.4779 144.2163    10
#                         as.IDate.default(ct) 197.6881 198.8312 208.2653 199.2498 226.9576 232.7463    10
#                                 as.IDate(ii) 101.8510 102.8230 113.2321 103.5843 127.8585 131.9020    10
#                                 as.IDate(nn) 146.1235 146.7726 150.7616 147.1008 148.0633 179.3617    10
#                                 as.IDate(ct) 152.6400 154.0221 157.7458 154.7143 158.7362 178.6318    10
microbenchmark(times = 10L,
               as.ITime.default(ii, origin = "1970-01-01"),
               as.ITime.default(nn, origin = "1970-01-01"),
               as.ITime.default(ct),
               as.ITime(ii),
               as.ITime(nn),
               as.ITime(ct))
# Unit: milliseconds
#                                         expr       min        lq      mean    median        uq       max neval
#  as.ITime.default(ii, origin = "1970-01-01") 1888.5241 1897.3788 1904.8904 1906.3941 1914.1901 1923.2183    10
#  as.ITime.default(nn, origin = "1970-01-01") 1886.1342 1888.4144 1901.7510 1898.2280 1911.8567 1931.2231    10
#                         as.ITime.default(ct)  883.3834  885.5955  899.8955  892.9906  915.1113  929.7641    10
#                                 as.ITime(ii)  276.6699  277.0064  283.8398  278.3184  287.2178  318.3477    10
#                                 as.ITime(nn)  180.4939  180.9974  183.0361  182.1190  183.8354  191.7105    10
#                                 as.ITime(ct)  189.3085  191.3544  207.5544  196.2207  202.4649  287.6540    10

[mattdowle]

This one is strange. It seems to merge with master ok but it fails its test with :

 Error in if (attr(x, "tzone") %in% c("UTC", "GMT")) as.ITime(unclass(x),  : 
    argument is of length zero

When I fix that locally to catch missing tzone, it still fails 3 of its tests with a date offset issue :

> test(1776.6, as.IDate.default(p), as.IDate(p))
Running test id 1776.6     Test 1776.6 ran without errors but failed check that x equals y:
> x = as.IDate.default(p) 
First 6 of 1000000 :[1] "2015-01-16" "2015-02-24" "2015-05-09" "2015-09-08" "2014-12-24" "2015-09-04"
> y = as.IDate(p) 
First 6 of 1000000 :[1] "2015-01-17" "2015-02-25" "2015-05-09" "2015-09-08" "2014-12-24" "2015-09-05"
Mean relative difference: 6.046997e-05