[optimization] Reduce RAM/VRAM traffic in graph.py by grouping nodes to execute by model type as well

[optimization] Reduce RAM/VRAM traffic in graph.py by grouping nodes to execute by model type as well #4926