Currently grib_encode_size_t / grib_decode_size_t are defined and
especially on windows where size_t is 64 bit,
encode_double_array and decode_double_array use them.

So especially for 32 bit linux system, define uint64_t type grib_encode /
grib_decode and use them instead of size_t type ones. This method is
more portable.

The proposal patch fixes the following test failure on Linux 32 bit:

#  35 - eccodes_t_bufr_ecc-1290