Move recursive loading of DataContainer to HDF5Content

[pmrv]

This will allow users and internal code to use job.content[...] if they what they are looking for is saved in HDF5 storage and obviate the need to rely on job[...] which may be slow because it is also searching the compressed job files.

I will need some logic like this to fix the slow loading of pyiron_atomistics jobs.

[pmrv]

The last commit should actually fix pyiron/pyiron_atomistics#1348 and pyiron/pyiron_atomistics#1350, but I'll probably add them in anyway for cleanliness.

As of this PR one can access various things in a job like this:

• job.project_hdf5 access to the raw HDF5 layout
• job.content access to HDF, but try to load any DataContainer along the given path (this was previously exclusive to job[...]) this could be extended to other objects that can be to_object()ed.
• job.files access to files
• job[...] will try to act first like content, then like files if a string without slashes was passed and finally will try to look up child jobs.

[jan-janssen]

Looks good to me. Still we should think about adding a benchmark which times typical job[item] queries and make sure we do not merge changes in the future which slow down.