Conversion between coe and rv is not transitive

[astrojuanlu]

This is a consequence on my indecision about using the semimajor axis or the semilatus rectum. To avoid singularities and problems (see #37) I should probably move to using the semilatus rectum definitively.

If I make this change, probably the special function introduced in #38 won't be needed anymore.

[astrojuanlu]

My indecision still holds, so I'm reopening this issue.

I find myself never dealing with parabolic orbits, yet I introduced this change and I find it extremely annoying for the remaining 100 % of my use cases. I'm thinking of reverting it once and for all.