utv_sum_tangent_stack
=====================

.. py:function:: t3toolbox.backend.utv_operations.utv_sum_tangent_stack(variations_data, n_tangent, axis = None)

   .. code-block:: python

      def utv_sum_tangent_stack(
              variations_data,         # variations .data, supercore stack = K + C
              n_tangent:   int,        # |K|
              axis:        typ.Optional[int] = None,  # 0-based index WITHIN K (None = the whole tangent stack)
      ):  # -> variations .data with the summed K axes removed (stack = C, or K-with-one-axis-removed)


   Sum the variations over the tangent stack ``K`` (a batch of tangents at one frame -> their sum;
   corewise == the tangent sum, by linearity). The frame stack ``C`` is preserved.

   The supercores sum via ``xnp``; the masks **OR** over the same axes (host ``np``). Because a ``K`` stack
   shares one frame, its masks are constant along ``K``, so the OR is a no-op (the summed tangent carries
   the frame's gauge masks) -- but it is the correct reduction in general. ``axis`` indexes within ``K``.