sumsq_over_probes
=================

.. py:function:: t3toolbox.backend.fitting.sumsq_over_probes(zz, n_w)

   .. code-block:: python

      def sumsq_over_probes(
              zz:     typ.Union[typ.Sequence[NDArray], NDArray],  # ragged len=d (elm W+C+(Ni,)) OR packed (d,)+W+C+(N,)
              n_w:    int,                                         # number of leading sample-stack (W) axes
      ) -> NDArray:                           # sum of squares over W and the free mode, summed over d, keep C


   The ``‖·‖²`` reduction for the vector-output ``probe`` kind: sum over the leading ``n_w`` sample
   axes and the trailing free mode, keeping the frame stack ``C``, summed over the ``d`` probes. **Mirrors**
   ``zz``'s packedness: a ragged ``len=d`` sequence loops over ``d`` (each ``z_i`` is ``W+C+(Ni,)``); a
   packed ``(d,)+W+C+(N,)`` array sums ``d`` + ``W`` + the padded mode ``N`` in one op (the free-mode
   padding is a zeroed prefix, so it contributes nothing -- the packed inner-loop path).