Calculating bias/RMSE with multiple obtypes; x-axis limits when obcount is 0

[BenjaminBlake-NOAA]

Two proposed fixes for the plot_gsi_stat_exp.py script are listed below:

Calculating the bias/RMSE values in the script only works if you select one obtype. When reading in multiple obtypes, the mean (bias and RMSE) and sum (ob count) calculations are incorrect because the script still calculates the values using one obtype.

If there are no observations in a given vertical layer (the count is 0), then dividing by 0 to get the bias/RMSE will result in NaNs. When there are NaNs in the arrays, the x-axis limits for the bias/RMSE plots don't get set properly. Some code could be added to ignore the NaNs in the min/max calculation for the x-axis limits.