Build multi-architecture docker images using docker buildx

[jay-m-dev]

We need to build all docker images: dbmongo, machine, lab, unit_tester, int_tester using buildx, which allows targeting multiple architectures from a single machine (instead of having to build on the target system)

So far, all python packages, mongo db and nodejs are being updated to latest versions (see issue #481). scikit-learn, pandas, numpy and surprise v1.1.1.1 will still need to be updated to versions that have arm64 available from PyPi or piwheels.

A workaround for the missing wheels is to have the compiled wheels in the repo until we can break from completely from this dependency.

[jay-m-dev]

All packages have been updated to compatible versions for both x64 and arm64. Building docker images on arm64 takes a long time, since all requirements' wheels are built during this process. To speed up the docker build process we will need to prebuild the updated versions of wheels for all requirements.
As of right now, x64 and arm64 docker images need to be built on systems of the corresponding architecture. But we are very close to making this work with docker buildx.

[jay-m-dev]

Reopening issue, it was closed by branch merge. Removed branch link. We still need to setup and make docker buildx work.

[jay-m-dev]

Docker buildx is now being used to build multi-architecture machines. This is part of the release process, images are pushed directly to docker hub for release. Docker-compose can still be used as usual for local development.