Rewrite and simplify ReturnFromMoon

Reject the Sims-Flanaghan-inspired approach and just use two
Lambert phases, switch to SLP solver, improve the sliding windows
for the initial guess.

Signed-off-by: Lamont Granquist <lamont@scriptkiddie.org>

MechJeb2/OrbitalManeuverCalculator.cs

@@ -362,10 +362,8 @@ public static (Vector3d dv, double dt) DeltaVAndTimeForMoonReturnEjection(Orbit
             double moonSOI = moon.sphereOfInfluence;
             (V3 r0, V3 v0) = o.RightHandedStateVectorsAtUT(ut);
 
-            double dtmin = o.eccentricity >= 1 ? 0 : double.NegativeInfinity;
-
             (V3 dv, double dt, double newPeR) = solver.NextManeuver(primary.gravParameter, moon.gravParameter, moonR0,
-                moonV0, moonSOI, r0, v0, targetPrimaryRadius, 0, dtmin);
+                moonV0, moonSOI, r0, v0, targetPrimaryRadius, 0);
 
             Debug.Log($"Solved PeR from calcluator: {newPeR}");