Fix broken save_audio

[flyingleafe]

Commit 2046b55 introduced a silly yet horrible regression, completely breaking down save_audio method:

• It stopped working in the first place for MixedCuts, since they do not have recording field, so fastcopy fails there;
• It stopped really working for DataCuts as well, since fastcopy preserved the original start times of the cut, while the underlying recording had changed, and so every such cut became invalid after save_audio (except maybe for the one starting from the very beginning of the original recording.

I do not have time today to add regression test for this, maybe tomorrow.

This amounted to my almost entire working day being completely puzzled about why my data preparation pipeline got completely broken after I just changed the underlying data.... been blaming the data itself initially, not the master update.

(Basically, I did some .cut_into_windows + .padding + .filter and then .save_audios)

[pzelasko]

Good catch, sorry for that! I agree it should have a better test coverage. I'll merge it as-is to fix the broken behavior. Would you be willing to make a follow-up PR with tests?

[flyingleafe]

@pzelasko If I have some spare time tomorrow, will do the test