ci: improve runtime by disabling SMP on db820c to work around lack of cpufreq
- gles2 7.5min -> 13min
- gles3 17min -> 3min (and 2.5x the tests run)
- gles31 10.5min -> 6min (and 10x the tests run)
I'm not sure what's up with gles2 getting slower when gles3/31 got faster, but this reduces our total runtime so we should go with it until upstream gets cpufreq.