Rethink pipeline graphs with `when: on_failure`
Description
The whole when: on_failure
thing really messes up pipelines. I hate seeing a green after a red. As a user glancing at the pipeline, it just makes no sense. If it weren't a straight line, and the failure vectored off like a flow chart hitting a decision diamond and then picking a failure path, it might make sense. But continuing on in a straight line on failure just feels wrong. I'm not saying we need to solve that with graphic design. Part of the answer is to move slack notifications to a more feature-full service that is outside the pipeline. But there will always be edge cases. This issue is to discuss any potential changes to the pipeline graph itself to help.
Proposal
Links / references
/cc @dimitrieh