More details are in #1113.

TLDR:

• avoid collecting the same position in the text when the note label is the same. So for example if we have This note1, and this note2, but back to the first note1, we would collect twice the offset of the first 1 label.
• update the labels2notes so that we use the identifier instead.
• fix a NPE when the note tokenization are empty, the processShort return a null, so we skip such note