Isolate docker build environments

[ImogenBits]

this addresses #64 by isolating the build environments of each image from each other. To do this I archive the newly created image to disk. This comes at a significant performance hit so I also introduced the --unsafe_build CLI flag to have quicker local debugging runs.

[Benezivas]

If I understand it correctly, we may raise an IndexError here if the very first generator that is being built is not built successfully, as image_archives is still empty at this point.

[ImogenBits]

image_archives doesn't just contain the paths to generator archives but also solvers, since they are being built first there never will be a generator that is being built first. the intention is to clean up the solver of the team whose generator failed, not to remove another generator.

[Benezivas]

Thank you for the pull request, could you please look into the possible exception I commented on and reduce the complexity of the _build method slightly such that the checks run through? I think good candidates for pruning are the first two checks of the method regarding the list of teams.

[ImogenBits]

idk why github isnt running the checks but the linting issues are gone

[Benezivas]

I had to remove some type hinting as python 3.6 does not implement all of them yet. I like the new option of an --unsafe_build.

Testing shows that we indeed have a performance hit during execution, but the --unsafe_build mitigates at least a little bit of that.
Maybe we can think about optionally loading prebuilt images from users instead of building from directories for version 4.0.0. Since we are now storing and loading images anyway during setup, this change should not require a lot of additional effort. This should also speed up local development by quite a margin, since users can then use dockers caching strategy in order to avoid rebuilding possibly expensive initial setup steps.

Thank you for your work. I will publish this PR together with the encoding fix as a new Subversion.

[ImogenBits]

Maybe we can think about optionally loading prebuilt images from users instead of building from directories for version 4.0.0.

we essentially already do that, if there already is an image build from that file in the cache docker won't actually build a new one, without the new safe build the battle startup time is pretty marginal already.