This was done because the random-number tests that chose
parameters were sometimes choosing parameters that failed
by roundoff.

Now this passes with --repeat-until-fail 100000

The particular changes:
 - FocallyLiftedMapHelpers::try_scale_factor now has a choice when
   setting up the quadratic equation to solve for lambda.  It can
   choose to solve for lambda-1 (the default, and the previous
   behavior), or it can solve directly for lambda. The difference is
   only roundoff but makes a difference in roundoff tests.
 - Accordingly, FocallyLiftedSide, when solving for lambda_tide, now
   sets up the quadratic to solve for lambda_tilde instead of
   lambda_tilde-1.
 - We separated out a particular use case: projection point exactly on
   one of the z-planes when sphere_two > sphere_one. This use case is
   now tested tested separately, and has different allowed parameters.
   Changes in the allowed parameters for the (now 3) different cases:
 - For sphere_two > sphere_one and projection point on the z-plane:
    - center_one must have the same x and y coordinates as center_two.
    - Rho must be zero (to roundoff)
    - The z-planes must be farther from the edge of sphere_one.
    - Radius_two is allowed to be larger, and radius_one cannot be
      too small.
 - For sphere_two > sphere_one and projection point not on the z-plane:
    - rho must be smaller than previously allowed.
    - radius_one cannot be too small and radius_two cannot be too large.
    - proj_center must be sufficiently far from the z-planes.
 - For sphere_two < sphere_one
    - Keep projection_point far enough from surface of sphere_two
    - Keep sphere_one and sphere_two sufficiently far apart.
    - Keep the z-planes sufficiently far from the centers or edges of
      sphere_one.