CI: Consider using more AC_REPLACE_FUNCS()

[jimklimov]

NUT sources include a number of fallback implementations of certain methods that not all target platforms provide on their own.

Ways to integrate those differ:

• a few are listed in configure.ac in the AC_REPLACE_FUNCS() method, causing autotools to generate the needed goals for missing code, and are listed as LTLIBOBJS macro in that build circumstance
• many cases are listed in our Makefile.am explicitly, as a pattern like

if !HAVE_SOME_METHOD
  something_SOURCES += some_method.c
endif !HAVE_SOME_METHOD

• possibly other tricks are also used (primarily in test code optionally including C files like headers, but that may be or not be not too interesting for this clean-up)

This issue is about investigating more about this, and perhaps simplify recipes where we needlessly re-invented a wheel we are already using elsewhere.

From work done in #2825/#2828 we know that LTLIBOBJS should be used carefully, as they are tacked into our helper libraries (mostly under common/) and it is very possible to end up including two libraries with competing (albeit identical) copies of the object code, or a library and object explicitly (when building a library).

It may make sense to have a further helper library made of only LTLIBOBJS (and maybe some dummy object so it is never empty) that all others can depend on, ensuring a single implementation as far as linker is concerned. This may be (or not be) coupled with #2097 to provide the strings from nut_version.h as that "dummy object".

Maybe all other Makefiles have to be updated, so that library is linked into final programs and published libraries (static/shared) explicitly and on same level as other libcommon*.la helpers, and not through them (complicating the dependency and rebuild tree, and possibly re-introducing those linker conflicts).