Use double precision for time bounds

[blimlim]

Closes #96.

Time bounds precision:

The time variable uses double precision while its bounds use single precision. When the time is large relative to the origin, inaccurate bounds can be written to the output files for hourly data. E.g. for ESM1.6:

$ ncdump -h iceh-1hourly-mean_0272-01.nc 
...
 time = "1850-01-01 02", "1850-01-01 03", "1850-01-01 04", "1850-01-01 05", 


$  ncdump -t -v time_bounds iceh-1hourly-mean_1850-01.nc 
 time_bounds =
  "1850-01-01", "1850-01-01 01:30",
  "1850-01-01 01:30", "1850-01-01 03",
  "1850-01-01 03", "1850-01-01 04:30",
  "1850-01-01 04:30", "1850-01-01 04:30",

This PR swaps the time bounds variables to use double precision. With the changes, we get:

 time_bounds =
  "1850-01-01", "1850-01-01 02",
  "1850-01-01 02", "1850-01-01 03",
  "1850-01-01 03", "1850-01-01 04",

Using the release-1deg_jra55_iaf branch for OM2 there aren't any immediate differences in the bounds, as the time origin is much closer to the data:

Without changes

time_bounds =
  "1958-01-01", "1958-01-01 03",
  "1958-01-01 03", "1958-01-01 04:30",
  "1958-01-01 04:30", "1958-01-01 06",

With changes

 time_bounds =
  "1958-01-01", "1958-01-01 03",
  "1958-01-01 03", "1958-01-01 04:30",
  "1958-01-01 04:30", "1958-01-01 06",

Bounds on instantaneous data

For the netCDF io, the lower time bound for single timestep data is also off:

$ ncdump -t -v time_bounds iceh-1-mean_1850-01.nc 
time_bounds =
  "0001-01-01", "1850-01-01 01:30",
  "0001-01-01", "1850-01-01 01:30",
  "0001-01-01", "1850-01-01 03",
  "0001-01-01", "1850-01-01 04:30",
  "0001-01-01", "1850-01-01 04:30",

The io_pio/ice_history_write.F90 code already omits time bounds for histfreq(ns) == 1. This PR copies this change over to io_netcdf/ice_history_write.F90.

Unaddressed issues

There are still a couple of odd things with the hourly data:

• It seems to skip the first hour:

   ncdump -h iceh-1hourly-mean_1850-01.nc
   time = "1850-01-01 02", "1850-01-01 03", "1850-01-01 04", "1850-01-01 05", 
  

  which might be related to this line. I'm not sure whether we'd like to address this here.
• Each run produces two output files for hourly output. The second only contains the single timestep at the boundary of the month/year:

  $ ls archive/output000/ice
  i eh-1hourly-mean_1850-01.nc  iceh-1hourly-mean_1850-02.nc  iceh-1monthly-mean_1850-01.nc iceh-1daily-mean_1850-01.nc  iceh-1-mean_1850-01.nc 
  

  Though I'm not sure how much we can do about this.