remove trained models

[satra]

Removing trained-models from the docker container, as a result also git-annex and datalad

Some notes on debugging:

• the tensorflow images are built using jammy (22.04)
• git-annex 10.2306 is available from neurodebian for jammy
• datalad get results in an error for one file from OSF (kwyk/bvwn_multi_prior/weights/variables.index) - the md5 object for that file does exist on osf.
• the same issue could exist in nobrainer zoo cli (i would suggest using mamba/conda for that image and installing datalad from conda-forge).
• the current trained-models repo is almost 1G, hence avoiding it can be beneficial.