Only process 1 graph request at a time for a given run

[lewisbirks]

An initial look into addressing #886, for #862 the initial computationally expensive part is creating the PipelineNodeTreeScanner and as such I have created a way of ensuring that all incoming requests use the same parent PipelineNodeGraphAdapter.
This required some additional synchronisation code to ensure that concurrent modifications aren't made to the underlying data.

Testing done

curl --parallel --parallel-immediate --parallel-max 30 --config urls.txt

where urls.txt contains about 90 entries that look like

url = http://localhost:8080/jenkins/job/demo/69/pipeline-overview/tree
output = ./results/tree/01.json
url = http://localhost:8080/jenkins/job/demo/69/pipeline-overview/allSteps
output = ./results/steps/01.json

These files can then be checked to all be the same using a diff

for f1 in ./results/steps/*; do
    for f2 in ./results/steps/*; do
        if [[ "$f2" > "$f1" ]]; then
            if ! diff -B "$f1" "$f2" ; then
                echo "Files $f1 and $f2 differ"
            fi
        fi
    done
done
echo "Checks complete"

Submitter checklist

• Make sure you are opening from a topic/feature/bugfix branch (right side) and not your main branch!
• Ensure that the pull request title represents the desired changelog entry
• Please describe what you did
• Link to relevant issues in GitHub or Jira
• Link to relevant pull requests, esp. upstream and downstream changes
• Ensure you have provided tests that demonstrate the feature works or the issue is fixed

[lewisbirks]

Thinking about it, a better approach might be to wrap the XXXApi objects instead as that would remove the need to worry about mutation. However this does still present some issues. The current way of working expects all of these to be single use objects, as seen with the need to deal the FlowNodeWrapper being mutated. If we cached at an Api object level we would want to somehow use the same PipelineNodeTreeScanner or PipelineNodeGraphAdapter between api objects as that is what is doing a lot of the heavy computation

[timja]

LGTM, tested locally with curl running multiple requests at the same time and can see only one gets 'built'.