Consider testing that objective returns a vector

[jmlarson1]

Both pounders and manifold sampling are designed to utilize vector output.

Because this is not typical of most optimization methods; users might not return a vector.

If the objective returns a scalar, pounders.m and pounders.py will not error. When given a scalar, both will store m copies of that value for each row in its Fvec array. Pounders will converge, but likely (much) more slowly than if vector of m outputs was returned.