Problem 780

On both Linux and Sun, I encountered program aborts which I traced back to
a NAN-producing operation in G4BraggModel::SampleSecondaries and
G4BetheBlochModel::SampleSecondaries.  The problem is in the lines:

G4double cost = deltaKinEnergy * (energy + electron_mass_c2) /
                                   (deltaMomentum * totMomentum);
G4double sint = sqrt(1.0 - cost*cost);

cost is occaisionally > 1 by a small amount, causing the argument of sqrt to be
negative.  My temporary fix was to protect the calculation of cost by taking
the std::min of the calculated value and 1.0.

The same problem also occurs in G4BraggIonModel and G4MollerBhabhaModel.