Fix size() for large variables >= 2^31 values

[Dave-Allured]

Refer to #476. This PR might resolve the issue.

This is a straightforward PR, but it is untested. Please run this through CI, or let me know how I can enable CI on my personal fork of netcdf-fortran.

This PR supports large array variables with aggregate size (total number of values [not bytes]) exceeding the original code limit of MAX_INT (2^31 - 1).

The original library code created temp arrays like this. The original fortran 90 size() intrinsic function returned only 32-bit integers. This overflowed and failed whenever the total number of elements (product of multiple dimensions) exceeded MAX_INT:

allocate(defaultIntArray(size(values)))

Fortran 2003+ allows 64-bit integers to be optionally specified for the size() intrinsic. The array limit now becomes MAX_INT64 (2^63 - 1) total number of values:

allocate(defaultIntArray(size(values,kind=EightByteInt)))

This PR fixes all 28 instances of this allocate construction, across three different source files in the fortran/ source directory.

This PR will require a minimum standard of fortran 2003 to build netcdf-fortran. I do not know whether the current overall requirement has already reached fortran 2003.