Handle sub-second representation error

According to ?`%/%`:

'%%' and 'x %/% y' can be used for non-integer 'y', e.g. '1 %/% 0.2',
but the results are subject to representation error and so may be
platform-dependent.

The example below shows one consequence of this.  The 'x' object has
an observation every 100ms, so we expect endpoints(x, "ms", 200) to
produce a sequence from 0 to 10 in steps of 2.  But the 4th observation
is 7 instead of 6, the cause of which you can see from the output of
sprintf():

  R> x <- .xts(1:10, .POSIXct(1.5e9 + 0:9 / 10))
  R> endpoints(x, "ms", 200)
  [1]  0  2  4  7  8 10
  R> sprintf("%0.0f", .index(x) %/% 1e-3)
   [1] "1500000000000" "1500000000099" "1500000000200" "1500000000300"
   [5] "1500000000400" "1500000000500" "1500000000599" "1500000000700"
   [9] "1500000000800" "1500000000900"

The output is more reasonable if we convert to milliseconds and
microseconds by multiplying instead of integer division by a fraction.

  R> sprintf("%0.0f", .index(x) * 1e3)
   [1] "1500000000000" "1500000000100" "1500000000200" "1500000000300"
   [5] "1500000000400" "1500000000500" "1500000000600" "1500000000700"
   [9] "1500000000800" "1500000000900"

We don't need to worry about values to the right of the decimal place,
because the cast to 64-bit integer in the C code will truncate the
double values toward zero (according to C99).

See #202.

R/endpoints.R

@@ -75,11 +75,13 @@ function(x,on='months',k=1) {
     },
     "ms" = ,
     "milliseconds" = {
-      .Call("endpoints", .index(x)%/%.001, 1L, k, addlast, PACKAGE='xts')
+      sec2ms <- .index(x) * 1e3
+      .Call("endpoints", sec2ms, 1L, k, addlast, PACKAGE='xts')
     },
     "us" = ,
     "microseconds" = {
-      .Call("endpoints", .index(x)%/%.000001, 1L, k, addlast, PACKAGE='xts')
+      sec2us <- .index(x) * 1e6
+      .Call("endpoints", sec2us, 1L, k, addlast, PACKAGE='xts')
     }
   )
 }

inst/unitTests/runit.endpoints.R

@@ -204,3 +204,15 @@ test.sub_second_resolution <- function() {
   ep <- endpoints(x, "ms", 250)
   checkIdentical(ep, c(0L, 3L, 5L, 6L))
 }
+
+# precision issues
+test.sub_second_resolution_exact <- function() {
+  x <- .xts(1:6, .POSIXct(0:5 / 10))
+  ep <- endpoints(x, "ms", 250)
+  checkIdentical(ep, c(0L, 3L, 5L, 6L))
+}
+test.sub_second_resolution_representation <- function() {
+  x <- .xts(1:10, .POSIXct(1.5e9 + 0:9 / 10))
+  ep <- endpoints(x, "ms", 200)
+  checkIdentical(ep, seq(0L, 10L, 2L))
+}