Made few changes to allow model-observation comparison statistics after the 4DVAR data assimilation. Many thanks to Andy and Brian for their suggestions.

- Modified is4dvar_ocean.h so the nonlinear model is run at the end after data assimilation, when using IS4DVAR. The nonlinear model is initialized with the estimated initial conditions from data assimilation. The nonlinear model is then interpolated at the observation locations for comparison statistics.
- Added a new routine Utility/stats_4dvar.F to compute model-observation comparisons. The following measurements are computed for each state variable:
1) Observations mean (obs_mean) and standard deviation (obs_std).

2) Model mean (model_mean) and standard deviation (model_std).

3) Model bias (model_bias).

4) Model-Observations standard deviation error (SDE).

5) Model-Observations cross-correlation (CC).

6) Model-Observations mean squared error (MSE).All these statistical quantities are written into 4DVAR output NetCDF file (MODname) and standard output file.

- Added stats_4dvar calls to all 4DVAR data assimilation drivers: is4dvar_ocean.h, s4dvar_ocean.h, w4dpsas_ocean.h, and w4dvar_ocean.h.
- Fixed an out-range index bound in arrayd A2d and A3d in the exact background error normalization coefficients routine normalization.F. This is a benign bug.
- Made changes to inp_par.F to insure that there is only one record in the restart file after the nonlinear model is run at the end of an assimilation cycle. The parameter nRST is set to ntimes. The restart file can be used as the first guess for the next assimilation file in sequential data assimilation.