Fixes issue #292
Grib2 uses packed (positive) integers, and the wgrib2 routines limit the length of integers to 25 bits. This is not a limitation for floating point numbers because the IEEE float format is limits the size of the mantissa to 24 bits (single precision). For most meteorological you only need 12-16 bits of precision. However, for precipitation people like high precision and routinely set the precision to 24 bits. The problem occurs with complex packing 3 which stores the deltas of the deltas. With an uncommon pattern of precip, the size of the scaled integers can increase 4 fold (+2 bits) which causes the scaled integers to exceed the 25 bit limit. This fix (1) updates the web documentation and (2) updates complex_pk.c which will not reduce the requested precision if necessary. Test for the fix was added.