Fix `redetect.py` to take into account speaker mapping gold standard

[mandlilaast]

Currently redetect.py may overwrite speaker mappings that were corrected using the gold standard in speaker-mapping issue #160 and its corresponding PR . As a result, running the script on the corpora can revert these manual corrections.

The script should be modified so that the gold-standard mappings are preserved (or prioritized) during redetection.

This should be implemented before updating the protocols with new data from the persons repository.

[ninpnin]

This might be an issue in terms of the gold standards train-test contamination. What do you think @MansMeg ?

[MansMeg]

Yes. Im not sure what gold standard is thought of here. But if it is the gold standard for estimation this should definitely not be done.