feat: use fgetpwent_r on linux

[reubeno]

Posting to get some conversation going around openness to using fgetpwent_r on the platforms that provide it (i.e., Linux).

Per docs, it's actually reentrant and could be used to implement a non-unsafe version of group and user enumeration on the platforms that support it. That somewhat begs the question of whether it's feasible (and/or a good idea) to provide functions that are only unsafe on some platforms. It would be great to get your feedback on whether you'd consider something in this direction.

(Aside: this macOS man page indicates that getpwent is thread-safe on that platform because they happen to place its internal buffer in a thread-specific structure. Would need to more properly validate that, though.)

[reubeno]

@gierens Any early feedback on this?

[gierens]

Hi @reubeno, thanks, and sorry for the late reply, I'm currently very busy, after next week I should have a bit more time, I'll try to look at it then :)

[reubeno]

@gierens Totally understood, thanks for the ack!

[gierens]

Hi again, this looks very interesting, also very clean and neatly documented. Thanks for that!

In my opinion this is a definite improvement. I'm not sure if there are any best practices regarding "target-dependent safety", and that surely is a public API change, but assuming there are no side-effects we don't see at the moment, I think, I actually prefer this streamlined version over having a separate interface, just for API consistency. I'm also pretty certain, that, if this would've been developed with fgetpwent_r originally, and then ported to targets without it, "target-dependent safety" is exactly what it would've ended up with.

Maybe just to entertain the thought, if one would want to introduce this slowly, we would need a separate Linux only AllUsersSafe or so and mark all_users as deprecated on Linux and provide a new all_users_safe as alternative, ... This does sound fairly messy. So, I'm still inclined to prefer your approach.