Also without lens retraction the camera stuck in 'rec' like in the last few sunrise runs.
->exec con:call_function(0xff099f1c) seems not to be the cause.
You should be able to do that using a slight modification of your existing rawopint_rs wrapper. I think all you need to change is to hardcode anything that's currently initialized from rs_opts.* and then run like
luar < file.lua
then download the images with imdl. If you delete the files using imdl -rm or imrm I'd generally suggest doing that in playback mode.
I uploaded the rawopint_rs.lua file to the camera and added in rec mode:
luar loadfile('A/CHDK/SCRIPTS/RAWOPINT_RS.LUA')()
and in play mode:
imdl -d="D:/sunrise/input/${name}" -rm
I deleted the PT_EraseAllFile code, since the files are now saved to the sd card.
The new setup works fine, like the other did if executed within a few minutes.
Tomorrow morning I will know if the camera survived the night.

What I did notice:
after calling
imdl -d="D:/sunrise/input/${name}" -rm
the photos were downloaded and deleted form the camera but I connected with chdkptp GUI and saw that in playback mode the counter of the photos was at 15/15 (this was the 3rd run with 5 photos shot each) in the second screenshot you can see that in files there are no files left. Maybe this could be related to the problem as well? when I click left and then right the camera does not display any thumbnail anymore but only Unidentified Image (also for the 15/15 image).
Also erasing all images with the canon menu didn't worked properly: at first the camera displayed no image. but after a run with 5 photos shot the counter of the photos was back at 20/20.
Maybe I should not switch to play anymore?

staying in rec I never hat this kind of problems.
Also maybe there is a way to "turn off" the sensor also in rec mode? like a native function call?
Although 50°C should not be a problem at all for an electronic component, I am still worried because the sun does not shine directly on the sensor at the moment and the permanent reading of the sensor, even though no pictures are taken, naturally causes the electronics to wear out faster. Also when the camera is in play mode and the sensor is turned off, the temperature in the sensor during the time lapse is not as high as when it is permanently reading the sensor.
I think saving on the camera is the better test, but if that solves it then trying filedummy would be worthwhile.
Looking at the CLI code
con:exec('rs_opts='..opts_s..' '..shootscript,{libs={'rs_shoot'},execinfo=execinfo})
it includes the rlibs when using -script, so you should just need to call rlib_shoot_filedummy() at the appropriate point in rawopint. I think I would put it just before hook_raw.continue()
Thanks, I will try that later on. I had a look into the lib but did not understand how does rlib_shoot_filedummy() know how to name the filedummy if no attribute was passed to the function?