Since I lack understanding of ConTeXt/luatex combo and have to pass data
for dictionaries containing other dictionaries to 3D model embedder I had
to abuse two existing parameters, "control" and "display" and rely on sane
lua code being put there.
And even the limited parameter correctness checks do not report problems,
they just silently fail to execute.
One more option to deal with 3D models is not to embed them as annotations
in the main PDF file, but to make a PDF with only 3D model in it and
attach it to the main PDF, linking to a clickable 2D image (like
attachfile2 LaTeX package). Donno if that functionality is in working
condition now and how to use it.