Qual tool cluster info - calculating the number of exec nodes on yarn should look at all executors over time when dynamic allocation on

[tgravescs]

Describe the bug
Currently the logic for calculating the number of nodes used on a YARN cluster only looks at the active executors. This is an ok approximation but with dynamic allocation this number could change a lot over the lifetime of an application.

Our logic:

 val activeExecInfo = executorIdToInfo.values.collect {
      case execInfo if execInfo.isActive => (execInfo.host, execInfo.totalCores)
    }

 activeHosts.toSet.size

We should be looking at the maximum number of executor going at any point in time and then calculating the number of nodes needed.