introduce switch_donor in adjust_mdots

[matthiasfabry]

When doing wind-mass transfer with accretion_j_dot, in the case when the direction of mass transfer changes, the accretor does not have its accreted_material_j set. This makes solve_omega_mix fail as j(k) == 0 for the added cells.

This introduces a check during adjust_mdots to prevent this from happening.
If the current donor ends up accreting while the current accretor ends up losing mass, a donor switch is warranted without adjusting anything related to the implicit MT solver. Likely that value is very low anyway (compared to the wind-mass loss rates).

The todo on the equatorial radius is valid, as a critically rotating star has its eq_radius about 17% larger than its volume equivalent radius, much larger than the inacurracy of the fit used for R_min. However, if the star is appreciably roche filling, the concept of an equatorial radius washes away. The point radius (largest extent on the x-axis connecting the stars) would likely be the best approximation here. I could supply those from my Roche integrations.