Using a matrix and job separation we can make the architectures compile
parallel to eachother, hopefully reducing the time required for builds
and also simplifying the process of building a single architecture.
A problem that we encountered is that with Python the resulting packed
tars are very large. Each release is in the order of tens of megabytes.
Using artifacts in our pipeline can easily make us surpass the maximum
size limit for free GitHub accounts (500 MB).
Because of this, we use the regular non-parallel pipeline for release
build. Releasing the version from the same job the build was performed
in allows us to directly access the build files instead of using
artifacts.
Separated release and MR pipelines.
a0ceeff014