treewide: compile fixes for native

[benpicco]

Contribution description

char is unsigned on ARM but signed on x86.
This lead to a few compile errors when adding the periph_gpio feature to native, because now code that was previously never build for x86 got hit by this difference.
The host compiler is also usually newer than the embedded one, which usually means it is stricter.

This makes offending chars explicitly unsigned and works around some pedantry.

Testing procedure

No logic changes. In fact, the binaries should stay the same.

Issues/PRs references

split off & needed for #12451
fixes #14667

[maribu]

See inline comments

[maribu]

Feel free to squash

[maribu]

See inline. You can directly squash in the style nitpick. Regarding the -Wno-pedantic: Does this only affect code out of our control?

[maribu]

Is it only me, or are the inline comments gone?

[benpicco]

I don't see it

[maribu]

OK, the comments was a another (void*) to (void *) that I initially overlooked, but you already fixed it. And the -Wno-pedantic indeed does only apply to the package and we cannot fix code outside of RIOTs tree.

So, both comments are addressed already.

[maribu]

ACK. This PR does three things:

• Use a different type name (uint8_t instead of char), which refers to the same type for all previously supported targets
• Use hex instead of binary representation (0x01 instead of 0xb00000001) of the same numbers
• Store a function pointer into an uint32_t instead of an void *

The first two changes do obviously not change anything in the generated binaries. The third change resulted in identical binaries being build of tests/pkg_semtech-loramac for the nucleo-f767zi according to elf_diff. Thus, no regressions are expected.