Consider unlocking dependency versions

I understand the need to lock versions of pandas, numpy, etc across the project, for primitives' and docker images' sake.

However I don't think version numbers need to be locked in the d3m package's setup.py.

Systems can use the d3m package outside of the primitives image context, for example the Datamart systems, and having a locked version of numpy might mean that you can't use some packages that need a more recent numpy.