handling NaN fillvalue in runoff input data

If compiled with a flag `-ffpe-trap=invalid` and your runoff includes `NaN` as Fillvalue in the netCDF (like netCDF generated with xarray), the code stops with the compiler's message : `Program received signal SIGFPE: Floating-point exception - erroneous arithmetic operation` or similar to this.

The error happen at

`where ( abs(dummy - forc_data_in%fillvalue) < verySmall ) dummy = realMissing`

in read_runoff.f90.

Here fillvalue is NaN and the fortran trap `dummy - forc_data_in%fillvalue` as an invalid answer.

The code may need to handle NaN fillvalue by changing NaN to realMissing.