Move tensorflow and keras dendencies to conda from pip

[haimasree]

It is recommended, in an environment where dependencies are getting installed from both conda and pip - keep the pip dependencies to be limited to only the ones which are not available in conda. In kipoi model zoo, most of the model groups install tensorflow and keras from pip instead of conda. In this pr, I attempted to move tensorflow and keras installation to conda from pip for 12 model groups. I was not successful for some model groups such as KipoiSplice, labranchor, FactorNet etc. Older the tensorflow version, more difficult it seemed to resolve dependencies. I will update this information in the model contribution guide as well.

[haimasree]

Note: Although test_new_models did not pass, in this ci run all model groups passed tests. test_new_models tests the models sequentially leading to a time out since it is just not possible to test 24 models under 1 hour sequentially. Hence merging with admin priviledges