[Gridmix] Use high-RAM jobs' behavior/info in cluster overloading logic in STRESS mode

Currently, Gridmix in STRESS mode calculates the existing load on the cluster to decide on whether to load further or not. The number of map/reduce tasks of all jobs are considered in this calculation as "each task taking single slot". But with high RAM jobs, this is not true and so needs to fix this computation.