Add CentroidTuple3d op

[rimadoma]

Add an op which calculates the centroid of a collection of vectors (? extends org.scijava.vecmath.Tuple3d)

[ctrueden]

Thanks @rimadoma!

Before we merge this, I think we really need to decide how we are going to handle vectors, matrices, etc., as core data structures. I am rather against adding a vecmath dependency to ImageJ Ops core—though it is actually pretty lightweight, it is also GPL code, which is incompatible with the Ops permissive BSD license.

See also fiji/fiji#135

Anyway, this op might belong in its own imagej-ops-vecmath project or some such. What do others think?

[ctrueden]

Let's finally decide on matrix libraries at this 2017-12 hackathon! Then this PR can finally be resolved.

[rimadoma]

Yes, this thing could go with my other linear algebra ops.

There is already a dependency on apache.commons.math in imagej-ops, but I think there's call for mutability. If immutability is really needed, we could have wrappers on top of the mutable objects e.g. ImmutableVector3d

[rimadoma]

I resurrected this PR by updating it to use JOML, which is now the consensus linear algebra library (?). Suggestions about for the namespace of this op are welcome. Could start math/lin-alg/? I have at least one other op that would fit there. In any case, I expect eventually this'll better fit into scijava-ops.

[rimadoma]

If I wanted to create the same op method for a collection of Vector3d and Vector3f what would be the best way? Just centroid(Iterable<Vector3d>) will have the same erasure for any type. Also the Vector classes have no other common supertype than Externalizable.