SoM's analyzer currently lumps build failures in with test failures, all under the alert type of messages.AlertBuildFailure. 

Given that the tasks to remedy the two different scenarios are different, we should split these alerts out further in the UI. 

Build failures are also more severe than test failures in terms of how much work they block.

To start out we should add a messages.AlertTestFailure alert type and modify the analyzer code to set it properly based on whether or not infra/monitoring/analyzer/step/test_step.go was the step analyzer that identified the failure. (consider modifying the StepAnalyzer interface's Kind() function to return an actual messagess.AlertType value).