Too many CI failures; cache generated images

[Tieske]

Most of the CI failures (false negatives) are errors while building the test image;

• Pulling the Enterprise images from Bintray (95%)
• downloading packages during the build (5%)

If the generated images were cached, we could circumvent that, and also reduce the docker pulls. (possibly even speed up the tests)

Issues:

1. Caching is done based on test-matrix usually: since the Pongo version is not in the test matrix (Pongo is unversioned), we might end up running Pongo against an image build by an older version of Pongo.
2. common test matrix is running a single version, and hence a single image, how about multiple images?
3. How to make sure it works with multiple CI engines

Proposal:

• to cater for 1 above; pongo generates a cache file-name based of; a) it's own commit-id, b) standard named extension
• pongo load <dir>: if the filename exists in the cache-directory, the file is loaded (docker load?)
• pongo save <dir>: exports all pongo images to the file (docker save?)
• when saving, any other file in that dir with the pongo filename extension gets deleted (it would mean generated by a different Pongo version/commit-id), and since that is not covered by the CI matrix, Pongo should take care of this clean up

[Tieske]

Important: only Kong versioned images should be saved, nightly or other "moving" targets should never be written to the cache file

[Tieske]

to be fixed by #220