Annotations/enforce unicity

[LoannPeurey]

The purpose of the PR is to enforce that:

• we cannot have multiple lines in annotations.csv with the same set and annotation_filename
• A set cannot have overlapping annotation files. Meaning inside a specific set, it is not possible to annotate multiple times the same stretch of audio (bc then this stretch is counted multiple times in metrics and other analysis)

To achieve this, all the functions that interact with annotations.csv must make sure before writing to it that those cases don't exist,
this is especially relevant for importations that now will fail for every row that does not satisfy those conditions.

Previously, the errors encountered during importation would find themselves recorded into annotations.csv. I felt like this could be troublesome as the file is intended as a record of indexed annotation and should not serve as a logfile. My take was to output all the errors to a separate csv in extra but not too sure about this solution

[LoannPeurey]

I am also unsatisfied about the

ChildProject/ChildProject/utils.py

Line 180 in 6f6474f

def find_lines_involved_in_overlap(df: pd.DataFrame, onset_label: str = 'range_onset', offset_label:str = 'range_offset', recording_label: str= None, set_label: str = None):

function to handle correctly when we want to select by a number of other columns, I could not think immediately of a clean way to do it without eval() .

EDIT: changed to use eval, looks alright to me this way :)

[LoannPeurey]

This change also leads to annotations always being sorted by 'imported_at' values. However, a lot of lines have the same value for that, leading to annotations.csv being changed in the order inconsistently by every importation done. So write should always order before writing changes, I suggest by : 'imported_at' then 'set' then 'annotation_filename' . (set, annotation_filename) should be always unique, so this should be consistent.

And maybe add a test for that sorting?

[shuvayanti]

Everything looks good to me. But I see a merge conflict which probably need to be resolved.