Use ccache for CI

[Schneegans]

This may resolve #299

[coveralls]

Pull Request Test Coverage Report for Build 3675584560

• 0 of 0 changed or added relevant lines in 0 files are covered.
• 114 unchanged lines in 2 files lost coverage.
• Overall coverage decreased (-0.002%) to 2.932%

---

Files with Coverage Reduction	New Missed Lines	%
src/cs-gui/internal/JSHandler.hpp	2	0%
plugins/csp-web-api/src/Plugin.cpp	112	0%
<!--	Total:	114		-->

Totals
Change from base Build 3513272209:	-0.002%
Covered Lines:	491
Relevant Lines:	16745

---

💛 - Coveralls

[Schneegans]

Some preliminary results. The numbers are extracted from various builds and should only be considered as rough estimates.

build type	without caching	buildcache	sccache[^1]	ccache[^2]	clcache[^3]
gcc	24 min	8 min	7 min	4 min	-
gcc (pch + unity build)	12 min[^4]	7 min	6 min	12 min	-
clang	28 min	6 min	6 min	5 min	-
clang (pch + unity build)	14 min[^4]	7 min	6 min	5 min	-
msvc	> 60 min				?
msvc (ninja)					-
msvc (pch + unity build)	34 min	no caching[^5]	no caching[^5]	no caching[^5]	28 min
msvc (pch + unity build + ninja)	31 min	fails[^6]	32 min	23 min	-

[^1]: See #319 [^2]: See #320 [^3]: The clcache numbers I cannot find anymore as the logs for those runs do not exist anymore. But I found some ancient runs which took about this time. [^4]: Even after I switched to build type, some cache hits occurred, so the actual time could be a bit higher. [^5]: For some reason, I cannot get either of the tools to work with the standard Visual Studio target. With Ninja, they work though... [^6]: buildcache seems to have issues with precompiled headers.

[Schneegans]

So I think for now, we should stick to ccache. I'll put this into this PR here and close the others.

[Schneegans]

A significant part of the time the windows build takes is for downloading boost. Maybe we should consider setting up a container for building which already includes boost. This could speed up the process by more than five minutes.