libC kconfig: default to PICOLIBC for NATIVE_LIBRARY if POSIX_API or GETOPT

[aescolar]

When building with the POSIX_API we cannot use the host libC.
When using GETOPT it is similarly quite difficult for users
to use the host libC.
So to make it easier for users, let's just default to PICOBLIC
in those cases,
while we continue defaulting to the host library in other cases
so users can use the Linux APIs for whatever test functionality
they want.

[keith-packard]

Is there any case for which we prefer the host C library with native_sim? It seems like we should always use a C library that sits atop Zephyr OS APIs instead of the host C library which sits atop the host OS APIs?

[aescolar]

Is there any case for which we prefer the host C library with native_sim? It seems like we should always use a C library that sits atop Zephyr OS APIs instead of the host C library which sits atop the host OS APIs?

It is a trade off: The argument for always using picolibc is the uniformity with other targets, testing in native_sim being more comparable to testing in real targets, and avoiding developers calling by mistake into the host OS.
The arguments against:

1. Not all native_posix/sim drivers support building with embedded libraries yet: native drivers and backends not yet supported with embedded C libraries #60096 (This is the biggest limiting factor today)
2. Quite a few users know when they are calling into the host OS, and want to be able to do it easily (for testing/debugging purposes)
3. Increased build time (picolibc is built from the module)

When 1. is solved, I would still give the option to build with the host library for 2. and 3., but I would be quite inclined to default to PICOLIBC in native_sim.

[aescolar]

I just changed the commit to also default to picolibc for native library when getopt is set, as that config is quite tricky to get working with the host libC

[keith-packard]

This seems like a reasonable migration plan.

[jukkar]

LGTM

[cfriedt]

The nsi_foo() approach is interesting. I was going to approach that problem slightly differently, but that seems to be working well.