Propagate efficient `NDArray` and `staticTensor` scalar+tensor operations for `dynamicTensor`

[Iainmon]

Propagate efficient ndarray and staticTensor scalar+tensor operations such as

// lib/NDArray.chpl
operator +(a: ndarray(?rank,?eltType),c: ?scalarType): ndarray(rank,eltType)
        where isNumericType(scalarType) do
    return ndarray.scalarMapOp("+",a,c);

and

// lib/StaticTensor.chpl
operator +(a: staticTensor(?rank,?eltType),c: ?scalarType): staticTensor(rank,eltType)
        where isNumericType(scalarType) do
    return staticTensor.scalarMapOp("+",a,c : eltType);

up into the dynamicTensor world. Right now these operations are implemented as

// lib/DynamicTensor.chpl
operator +(a: dynamicTensor(?eltType),c: ?scalarType): dynamicTensor(eltType) 
        where isNumericType(scalarType) {
    for param rank in 1..maxRank {
        if a.checkRank(rank) {
            return (a.forceRank(rank) + c).eraseRank();
        }
    }
    halt("Could not determine rank in dynamicTensor + " + scalarType:string + ".");
}

Which is not the most efficient implementation, and should use the versions defined in lib/NDArray.chpl and lib/StaticTensor.chpl.

The solution is trivial, as the definitions in lib/DynamicTensor.chpl, which should use the definitions in lib/NDArray.chpl. But there should be tests verifying this relationship.