[JENKINS-49707] Allow node blocks from deleted pods to be retried (full version)

[jglick]

Downstream of jenkinsci/workflow-api-plugin#217, jenkinsci/workflow-durable-task-step-plugin#180, jenkinsci/workflow-basic-steps-plugin#195, and jenkinsci/pipeline-model-definition-plugin#533. Subsumes #1190; will split out two further PRs that separately add Declarative syntax support, and controller restart tests.

Example display in B.O. of a build (old syntax predating retry step)

podTemplate {
    node(POD_LABEL) {
        stage('beginning') {
            sh 'sleep 30'
            sh 'date > date.txt'
            stash name: 'stuff', includes: 'date.txt'
        }
    }
}
checkpoint 'middle' // for CloudBees CI, in case you are *not* using node retry
podTemplate {
    node(POD_LABEL) {
        stage('end') {
            unstash 'stuff'
            sh 'sleep 3h'
        }
    }
}
build job: JOB_NAME, wait: false

which was running in the end stage on one K8s cluster, backed up via Velero (including volume snapshots but excluding all Pods), restored into another cluster, and automatically fixed itself:

Also recovers from

kubectl delete --force=true --grace-period=0 pod …

[jglick]

Split into #1210 & #1211.