Fix Concatenate on different axes

• Keras Concatenate layer assumes axis=0 is the batch dimension, so if we are parsing that directly, we need to consider that, i.e.
  • for 2D concatenation, axis = 2 or axis = -1, means we use concatenate2d_1; otherwise we use concatenate2d_0
  • for 3D concatenation, axis = 3 or axis = -1, means we use concatenate3d_2; axis = 2 or axis = -2 means we use concatenate3d_1; otherwise we use concatenate3d_0
• Also fix shape inference