I let the other mentors do a complete review, I only point out one improvement that would be great to have

This is good start :).
I did some small comments. Please check with your mentors how you prefer to work: either work by diffs, or you create your own branch (do commits when you want) and once it's good create a diff of the set of commits you want to check.

2-3 potential modes:

• like in missing_letter, a list of words and an image, we need to select the good word.
• like in missing_letter, a list of images and a word, we need to select the good image corresponding to the word.
• (in a 2.5 dimension world, having an item in the middle, a person to place on the asked position?)

For now, at each level, we delete the existing elements (walls, holes...) and create them again for a new level.
It takes time to create and delete the elements (especially with slow computers/tablets). The aim is to have a pool of elements that will be created at start of the activity and that we can use to do the elements instead of creating all of them at each level.
We have to find a compromise between the number of elements created at start (no need to create 100 walls at the beginning, it will take too much time too) and the number of elements we may need for the next level to avoid to have to wait too much for the creation of the level.

Thank you,
merged in:
https://cgit.kde.org/gcompris.git/commit/?id=c7a267ace1e820d0a7b58e0c3d6fc7b87436a211
https://cgit.kde.org/gcompris.git/commit/?id=040e1d61ef2f5e02e8dcc9943c9a779572b2bed3
https://cgit.kde.org/gcompris.git/commit/?id=7d3b8ecf5c37552d03391d9427ef959de6c8a17d

Merged in with small changes: https://commits.kde.org/gcompris/863b5856e5dd0f96b7a19b11040cab5e538297d3