[benchmarking] profile generator metrics scraping improved resilience

[annapendleton]

Currently filters on scraping workload metrics relies on the following conditions:

• single pod-monitoring object with "$BACKEND-podmonitoring" across all clusters in the project -

  ai-on-gke/benchmarks/benchmark/tools/profile-generator/container/benchmark_serving.py

  Line 588 in ee94bdc

  # podmonitoring spec assumed to be named "$BACKEND-podmonitoring"

• single workload on the cluster tied to the above pod-monitoring object

We should make this filtering more specific to the desired workload in case there are additional pod monitoring objects in the same project, + if there are multiple pods that the pod monitoring object will relate to on the given cluster