`vector_sum_of_squares` should only rescale for floating-point types, and should not use Scalar(1.0) or Scalar(0.0) otherwise

[mhoemmen]

vector_sum_of_squares has the following issues:

1. It does rescaling for all types, not just floating-point types (and complex-of-floating-point types).
2. It attempts to construct Scalar(1.0) and Scalar(0.0) for all types.

See discussion here.