Reimplement test architecture for qemu unit tests

Right now we need a hardcoded workaround to get the right cflags and cpu settings to compile unit tests for qemu. Ideally we should not need that.

We used to just pull those from a test architecture that could be defined for each module saying what the right settings were to use in qemu for that module.

A good idea could be:
- have `unit_test_qemu` `retarget` and add a `test_qemu` architecture that could be defined in a `lib_*` module_settings.json file for the various supported architectures. This would contain the required cflags and cpu settings to make the test run on qemu independent of what the actual target settings are.