kernel kconfig: Do not enable STATIC_INIT_GNU w the old native_posix

[aescolar]

native_posix (unlike native_sim) or its brethren (NATIVE_APPLICATION) link together the "runner" code and the embedded code in one pass. This means that when CONFIG_STATIC_INIT_GNU is set, any host library code (like the llvm fuzzer) constructors will get postponed to the Zephyr initialization.
These libraries are unlikely to work if we do this. (the llvm fuzzer does not)

So let's instead not enable STATIC_INIT_GNU for these targets. This means possible embedded library constructors will continue to be picked during the link and be still called during the native runner initialization instead of during the Zephyr OS initialization as they were just before we introduced STATIC_INIT_GNU in
6e977ae

Note that native_posix will be deprecated shortly and its users are strongly encouraged to move to native_sim.

Fixes #75407
Related to #74682

[andyross]

Works for me.

[andyross]

Note the board swap seems to be working in SOF for me (thesofproject/sof#9280), so no rush on this. Whether it's worth merging for any other hypothetical native_posix users is not something I'll argue about. :)

[aescolar]

Note the board swap seems to be working in SOF for me (thesofproject/sof#9280), so no rush on this. Whether it's worth merging for any other hypothetical native_posix users is not something I'll argue about. :)

Thanks @andyross I think it will be best to merge this before 3.7. We still support native_posix, and just like the fuzzer failed, other host libraries may also.
And just like SOF was still using native_posix I expect some other projects will still be using it or whatever derivatives they built over the years.
This just reverts this constructor initialization behavior to what it was before, but only for native_posix and its old family of boards. So it should be safe enough (I run a full twister run on native_posix and this did not introduce any regression).