Nested extraction create name collision

With two nested archive containing the same filename:

dataset0.zip/
    train.zip/
        ...

dataset1.zip/
    train.zip/
        ...

The two inner filenames will be extracted on the same destination (extracted/ZIP.train.zip), creating a collision between the files:

path0 = dl_manager.download_and_extract('http://ds/dataset0.zip')
dl_manager.extract(path0 / 'train.zip')  # Extract to extracted/ZIP.train.zip

path1 = dl_manager.download_and_extract('http://ds/dataset1.zip')
dl_manager.extract(path1 / 'train.zip')  # Oups, skip extraction to reuse ZIP.train.zip from dataset0.zip

To avoid collisions, we could add the hash of the extracted path, like extract(path) extract at ZIP.hash(path).name instead of ZIP.name currently.

Start by locating the implementations of download_and_extract and extract, then trace how extracted paths are named and reused. Reproduce the two nested-archive cases from the issue and add coverage for identical inner filenames. Done means distinct source paths no longer collide while repeated extraction of the same path can still reuse its destination.