feat: v1.3.0

Reply to issue #17 about the corruption of the former splits. A notebook `corrupted_splits.ipynb` has been created to diagnostic the problem and explore a solution. The solution is to align the split on train and test splits from PEDES.

Those changes has been made to the final dataframe. In addition of the changes of the split, the SYSU samples has to be constructed again. This time, most of the samples are not from SYSU anymore.

Thus, the produced annotations are not by "base_dataset" anymore. We only have one base dataset, it's PEDES. Also, since we do not know how to format the annotations for SYSU evaluations (see issue #19) we do not produce the test annotations yet. This work has to be implemented in other releases of the version 1.3.x.

The unit tests has been updated accordingly. Also, new tests has been introduced to ensure that the final dataset is matching the new requirements.

There are still work to do about cleaning the cli and the source code with former annotations format (coco, update csv output). This will be done in version 1.4.x.

We also setup the dev tools for the package in the pyproject.toml file.