Jobs that are allow_failure:true shouldn't block the pipeline

Today, if a manual job is set to allow_failure:true it does not block the pipeline. This is great, and is useful in a number of scenarios.

There are also scenarios however where we want a job to always run, but still be allowed to fail and not block the larger pipeline. One example is generating performance measurements, if you are not setting minimum thresholds for passing there is no reason to block forward progress.

In these situations, it would be really handy to for non-manual jobs (the default) that are allowed to fail to not block the pipeline.