After thinking for a while, I came to this approximation being "good enough". On big scales, it will give too big numbers:

h(l) = tan(l)*l/2

where h is the elevation between two points due to the curvature, and l is the length between two points, measured as an angle in the tangens function.

So, let's calculate h(19 nm). In radians, 19 nm = 192pi/360/60 = 0.0055269. In metres, 19 nm = 19*1852m = 35188 m

Hence, h(19 nm) = 15.47m

Meaning that the light is supposed to be too faint to be seen long before it disappears due to the curvature. At least if my formula is right, which I believe it is.

Around 33 nm the light should become below the horizon.

A typical relatively big sailing boat will stretch 2 metres up from the water, meaning that at 7 nautic miles, only the mast should be seen. I will try to do some binocular observations next time I have the chance.