Support for grouped impacts #533
Description
It should be easy to allow the ranking function to fix multiple parameters at-a-time to evaluate grouped impacts with methods that involve parameter shifting (n.b. in covariance-based method, grouped impacts are trivially computed). Does it make sense to shift the auxiliary data for multiple parameters at a time? or is a quadrature sum of impacts the correct way? If quadrature sums are sufficient, then only NP shifting impacts will benefit from this functionality.