-Disable LDF "deep+" mode on AT90USB platforms. Appears not needed any longer (likely due to fix 7322df26ad/). Results in identical binary output and cuts compile time in half
-Disable Cartesio config from CircleCI (compile failure)
-Disable Geetech I3 Pro X GT2560 from CircleCI (compile failure)
-Enable EEPROM on Micromake example config (fix compile failure)
-Move FolgerTech/i3-2020 to AVR platform in CircleCI (fix CirculeCI build failure)
-Disable various examples failing to build in CircleCI
-Enable various examples no longer failing to build in CircleCI

Runs in parallel to TravisCI and runs *much* faster on same jobs.  Finishes all examples configs in <8 min on cache hit; <10 min on cache miss.  Cache miss is triggered on platformio.ini change. I'd suggest move to it entirely, but it is limited to just 1500 min of runtime per month and it isn't clear yet how much Marlin would need so suggest we implement in parallel like this, then observe.  If the example builds don't catch issues, and 1500 min turns out to be plenty, we can switch to implimenting standard arsenal of tests in CircleCI and ditch TravisCI to gain speed improvement (although I do like seeing the varied Travis avatars staring at me ;).

I agree that hand-crafted tests would be better, but until we have that, this can't hurt.

After merging, you'll need to add Marlin project to CircleCI at: https://circleci.com
Example results at: https://github.com/fiveangle/Marlin/commits/bf-circleci_config