Error reduction function

[ChrisRackauckas]

@devmotion 's Quorum Sensing problem is an interesting case where the "full error" is not what we really want, but a more refined function for how to calculate the error. This is because in that case we really want just the error of the third component. We can probably add an error function to the benchmarking tools that lets you choose how to collapse the difference.

[ChrisRackauckas]

The better way to do this is to just add a reduction function that allows the user to specify how to reduce the solution, and have that default to norm. Then you just just measure the error on specific indices, or measure errors of functions of the solution quite easily.