Integration with single pipeline steps and/or pipeline graph view

## Problem
What problem does this solve?
When a pipeline executes steps in parallel, the output in the regular protocol used by the plugin contains all the output of the parallel stages nested together. This can potentially make it hard for the AI to find the root cause or lead to wrong results.

## Solution
Describe your proposed solution:
The plugin could walk over the flownodes of the pipeline and check which ones have failed and obtain only the log of that step. In a parallel execution without failfast it might mean that more than step failed and there is not a single reason, so potentially several calls must be done.
The plugin could also inject into the console log of a step (e.g. a `sh` step that failed, `pipeline/212/execution/node/12/log/`)

## Alternatives
What alternative approaches have you considered?

## Additional Context
Ideally the plugin is also able to inject itself into the pipeline graph view plugin so that when lookin there are a failing step one could click a button there to analyze an error.