Azure Pipelines: When the build fails, we shouldn't run tests #1266
Description
Currently, if a PR/CI build fails, we attempt to run the tests anyways. This is doomed and provides no useful info.
To fix this, it appears that azure-pipelines.yml and azure-devops/run-build.yml should name: their steps, and use dependsOn: to establish dependencies. Then, the default condition of "Only when all previous dependencies have succeeded" should be effective.
I don't immediately know all of the dependencies that we should set up. The STL tests should definitely depend on the STL build. However, I suspect that "Publish Tests" should depend on the STL build, but should run regardless of whether the STL tests succeeded or failed. Similarly, clang-format and validate enforcement should depend on Build Support Tools, but they should run regardless of whether the other fails.
We probably don't need to set up literally every dependency imaginable - just ones that prevent huge wastes of time.