The latest advice - changing the font - did work out. Now the text itself comes out right into Calibre and conversion to EPUB as well.
However, this Calibre conversion messes up picture text location, creates a mysterious cover page from the first PDF page as an image and adds page numbers, which do not follow the flow of the e-book ... because they are the paperback page number locations. I think all of that is wrong - possibly it could be made right by conversion options.
Having said that - if I read now the exported PDF (from Scribus) into Amazon e-book directly - against the recommendation and warning about PDFs not possibly working - none of the problems are present that I laid out for Calibre.
So, the solution appears to be to change the font AND not use Calibre and .pub format in between ... just use the Scribus PDF. Not sure whether this is my end conclusion, but this is how it seems after initial testing.
The fonts that DO work, are OpenFormat -fonts and PostScript fonts. TrueType fonts, like Calibre, are giving the problems. This is perhaps against expectations.
I wish that Scribus would control ALL THE FONTS separately that can be used in a book. So that you would KNOW where every particular font comes from. Now, you have no idea because every application install and uninstall can change the fonts. Every operating system upgrade can change the fonts. It is even worse if you use Linux and Windows in turns - on Windows, you cannot possibly know where its fonts come from, because you have a bunch of applications that add to the fonts. When I moved from Windows to Ubuntu Studio, it was really difficult, because I had mistakenly used fonts that were Windows-application specific and do NOT appear on Linux. Thus, I would like to have ONLY those fonts that appear in Scribus itself ... then I could add to that in an controlled manner. Then the fonts would also be portable to some other computer running a different operating system, like win10, kubuntu, garudo or whatever.
In principle this is Reference Data -concept - all the data that an application output is based on, is managed separately with the application data itself (=the book in this case). You would archieve Reference Data (=fonts) together with the actual data (=the book).
This could be enchanged to all the other metadata as well, like Styles - but I guess with copy/export/import between documents you can achieve much of the same, but ideally, you would have unique Styles in a library and you could reference those in many documents.