We should remove oldest-supported-numpy whenever it is seen

[marcelotrevisani]

          We should remove `oldest-supported-numpy` whenever it is seen. That's used by wheels specifically to [configure builds with specific `numpy` versions]( https://github.com/scipy/oldest-supported-numpy/blob/23ece5a161d5957f5142991393d2ae689ee45b9d/setup.cfg#L17 ) (not sure why a wheel would use `oldest-supported-numpy` at runtime unless they are some kind of build tool)

In any event it doesn't make sense in the Conda case where we configure NumPy versions used in some conda_build_config.yaml or other variants

On the Conda side, we need only add numpy (unconstrained) to requirements/host and copy over any dependencies value of numpy to requirements/run

Originally posted by @jakirkham in #552 (comment)