Resolve asan testsuite issues

[matthiakl]

Re #5090
The testsuite reports some asan issues which should be addressed. See this log

Since leaks are suppressed all together in the CI, the testsuite should be run on a local machine to see, whether some leaks can be resolved as well.

[Noordfrees]

Interestingly the failure always seems to happen on just three configurations: g++ 5 and 6 under Bionic; both release builds plus the g++6 debug build.
I'm compiling Widelands in a Bionic container now to look at it…

---

It's reproducible with g++5 under bionic in the Casern test. Will investigate.

[Noordfrees]

The bug seems to be in boost::format (specifically I'm using version 1.65.1.0ubuntu1 there).

Rewriting this function:

widelands/src/graphic/style_manager.cc

Lines 385 to 390 in 9efa8ea

	std::string StyleManager::color_tag(const std::string& text, const RGBColor& color) {
	boost::format f("<font color=%s>%s</font>");
	f % color.hex_value();
	f % text;
	return f.str();
	}

to use plain string concatenation instead of boost::format fixes this crash, only to get a new crash a second later elsewhere (in as_mapobject_message) as we use boost::format a lot. Gotta love this library, we never have any sort of problems with it whatsoever, oh no…

#5081 gathers 99% of our boost::format calls in a single bformat interface function. From there it's only a small(ish) step to ripping it out completely, which I would consider the most desirable approach to fixing this bug.

---

Edit: To clarify, the testsuite with ASan shows multiple bugs, of which several but not all are boost-related. They should best be addressed one by one until all checks are green.