Periodical harvest of annotations-corrections pairs

[JeltevanBoheemen]

To implement #179, a mechanism should be created that looks for (new) triples of (for each transcript):

1. corrected treebank XML file
2. initial annotations SAF file
3. corrected annotations SAF file

In theory, 2 & 3 will only exist if a user manually corrected the transcript. This is not guaranteed however, because the same file could be reuploaded. A quick check by using SafReader results should determine the difference.

Periodical harvest could be performed using Celery periodic task, with a simple bash script to add the files to sasta-datasets in a structured format.