Consider let build fail when hpy-abi=universal and Py* symbols are discovered

[timfel]

This could happen when the stub loader is created, since the .so exists at that point. It would ensure that people don't accidentally build hpy universal binaries that only work because our implementations currently have cpyext (or are CPython)

[mattip]

Another option is to set up compilation so that any use of Python.h will fail. This could be done by setting the include macro and creating a "fake" Python.h. Maybe both methods are needed to actually prove the binary is truly universal.

[mattip]

I think this can be closed now that #371 is merged. Please reopen if I misunderstood.