- vector, tensor versions are defined component-wise
  to avoid intermediates.

  The base version uses the form "(1-t)*a + t*b" without any bounds
  checking (ie, will also extrapolate).