[DOC] Extended example which measure impact of `memsave_torch` using `memory_profiler`

[f-dangel]

Add an extended example which demonstrates the impact of our memory-saving layers on peak memory, add to docs.
A good starting point could be the run.py file in the visual_abstract folder.