ctx: update from upstream, vary optimization levels
build ctx with O3 for compositing and Os for rest, we run into instruction cache issues with too agressive inlining for all of ctx, in particular the core scan-converter does not like O3 and is even a bit unhappy about O2.