BatchFillerSparseTensor3D Hotspots

[coreyjadams]

Batch Filler Sparse Tensor 3D is showing two significant hotspots and one minor one. Testing with NEXT NEW sparse 3D simulation:

• 68% of the execution time of the process function is spent at get_data:
  

  larcv3/src/larcv3/app/threadio/BatchFillerSparseTensor3D.cxx

  Lines 107 to 108 in 0663596

  	auto const& voxel_data =
  	mgr.get_data<larcv3::EventSparseTensor3D>(_tensor3d_producer).as_vector().front();

• 20% of the execution time is spent calling ImageMeta::coordinate:
  

  larcv3/src/larcv3/app/threadio/BatchFillerSparseTensor3D.cxx

  Line 161 in 0663596

  auto coords = meta.coordinates(voxel.id());

• 7% of the time is spent calling the destructor for std::vector.

The get data call can not be improved directly here, but the coordinate call and destructor can be improved. Half of the call to coordinate is spent doing the modular division, and a faster way to do this would be nice.