perhaps the base could be shaped with slight concavity on the inner half, and steeper concavity for the outer half. If you are buying the glass (concave lens), you could buy a shallow lens and surround it with a steeper border, which you machine from any rigid material (metal or plastic). The top would circle when twirled, but soon settle in the central glass.
When it begins circling around the circumference, with the tip leaned on the external ridge surrounding the concavity, it continues in that way for a long time. The circumference is short, and the top travels it in very little time, so this circling motion is rapid, and the resulting centrifugal force does not permit to the top to settle at the center until the top spins slow enough, towards the end of the spin.
This doesn't happen if there is enough concavity in the base, for this reason I make it so.
I should give to only the central part of the base a shallow curvature, but I have to think how to do it; with my technique for making bases ,(it is polished steel, not glass), I tend to generate a uniform curvature in the spinning surfaces.
With high Cg, slight lean moves the Cg in the direction of lean and the side it leans towards becomes the heavy side, even if it was the light side before the lean.
I think you are right. Leaning amplifies, (low CG), or even changes, (high CG over a certain angle of leaning), the effect of weight of the heavy side on the direction of leaning. Still it is not clear what triggers the change of direction towards the 90 degrees position.