From 1.14, using linalg.lstsq without specifying rcond yields the following warning:

rcond parameter will change to the default of machine precision times max(M, N) where M and N are the input matrix dimensions.
To use the future default and silence this warning we advise to pass rcond=None, to keep using the old, explicitly pass rcond=-1.

Which seems to strongly indicate that rcond=None is the safe bet going further unless you have a reason not to.
However, the warning does not indicate that doing so will make such code break on any install which has still numpy <1.14 (which unfortunately with pip is not so infrequent).

TypeError: must be real number, not NoneType

This is a minor gripe and proper regression testing on the user end should prevent big problems, but I still think it would be nice to be more diligent when introducing those types of changes.