gh-127081: use re-entrant variants of get{proto,serv}by{name,port}

[duaneg]

Add configure tests and defines for getservbyname_r, getservbyport_r, and getprotobyname_r. Use these if available, otherwise fallback to the thread-unsafe variants.

Add a unit test to exercise getprotobyname, which is currently untested.

TODO:

• Are there any platforms which define the unsafe variants but not the re-entrant ones? If not we can simplify the #ifdef hell somewhat.
• Do the re-entrant functions have the same signature on all platforms?
• These changes follow the existing code's practice: allocate a fixed-size (and overly large) buffer, and don't properly handle the error case if it is too small. Should this be fixed? If so should existing code also be fixed?

• Issue: Thread-unsafe libc functions #127081

[erlend-aasland]

Converted to draft since the author still has a TODO list for this PR. Moreover, error handling is missing, so this is a work in progress.

[duaneg]

With regards the TODO questions, I need some guidance from the maintainers...

1. I suspect there aren't any (supported) platforms that define the original version but not the re-entrant one, so we can simplify configure and the #ifdef clutter by only checking for/using the latter. If the maintainers consider that a potentially worthwhile improvement I'll push an updated patch to be confirmed or otherwise via the buildbots. If it is considered too risky and/or not worth the code churn then fair enough: we can just drop this TODO.
2. Triggering buildbot tests for this PR would answer the second question. That requires a core dev, though. Also, if we are going to try the #ifdef changes above I suppose it would be better to wait until after that.
3. It seems like the answer to "should this be fixed" is "yes", which is great! 🙂 As mentioned, I was following the pattern of the existing code (in socket_gethostname, socket_gethostbyname_ex and socket_gethostbyaddr) that doesn't handle buffer overflows gracefully. Personally, I'd prefer to fix the existing code and have everything consistent, but I'd understand if the maintainers would prefer to minimise the scope of change. What approach would you like me to take here? Just add buffer overflow handling to the new code or also fix up the existing code?