psVectorStats raises errors in too many places

[eugene]

psVectorStats raises errors for various situations that really represent a limitation of the data. For example, robustMedian will raise an error if there are no valid elements. these situations should be handled by setting the result to nan or appropriate saturation values, and maybe raising a warning.

I've modified psStats.c to avoid the error in the appropriate cases in my copy of the code. This needs to be made consistent with the uses in the rest of the IPP: places that relied only on a false return value should also check for NAN results (or we should always return false if the desired value is NAN, even if that does not result in an error being raised).