Fix some problems building recent NUT on Solaris 8

[jimklimov]

That was an interesting trip down the memory lane... see #1736 (and #1737 to an extent - not addressed much here except handling the situation in tests)

Now let CI chew at this, whether dumbed-down shell and perl scripts for syntax of the past millenium are compatible with all the new kids on the block? ;)

[jimklimov]

TODO: need a better test for availability of round() (and some others seeked nearby in math.h)

[jimklimov]

Oddly, several platforms failed a C++ test that this did not intentionally touch on:

cpputest-client.cpp:320:Assertion
Test name: nut::NutActiveClientTest::test_auth_user
assertion failed
- Expression: !s1.empty()
- Did not expect empty value here

Failures !!!
Run: 5   Failure total: 1   Failures: 1   Errors: 0
D: Got to the end of test suite with code 'false'
[ERROR] cppnit complained, check above

[jimklimov]

As another clue, it seems to fail this way in distcheck - so maybe something misbehaves for out-of-tree builds/tests?..

UPDATE: just building out of tree (mkdir -p .build ; cd .build ; ../ci_build.sh) also did not fail for main build in current directory, but also failed the distcheck.

Practically reproducibility is constrained to systems capable of building C++11 and CPPUNIT tests (Solaris 8 is not among these), so gcc-4.9+ and clang/icc; both exposed the problem, on openbsd VM and debian-ish containers...

[jimklimov]

In the end, somehow sed -i avoidance was complicit. Another way to skin the cat worked though...