Allowing node_pipeline to deal with return_scaled

[yger]

As mentionned in #2387 this PR fixes the problem of return_scaled within the node pipeline. Now, everything will be handled with the scaling, and while it might be slightly slower (marginally), it will avoid issues as observed in #2346 . Plus, all the templates obtained via internal spike sorting pipelines will be in appropriate units

[yger]

Tests are failing because the problem is more general than I thought, and need some discussion. I do not get why the signature for recording and recordingsegment of the get_traces() should be different. This is at the root of all issues, since some low-level functions are only relying on traces from segments. However, the discrepancy here prevents us to use scaling consistently. We should discuss if this is a built-in features, or if behaviour should be harmonized @samuelgarcia

[yger]

I made a solution keeping intact get_chunk_with_margin(), and simply adds an internal get_traces_with_margin() used in node_pipeline. This allows pipelines to deal with scaled data (better for peak amplitudes)