FVT support for more runtimes/tests

[rafvasq]

At the moment, FVT workflow is limited to running only tests for ~3 runtimes and as of recently ModelMesh supports 4.

Other than increasing the resources on the FVT cluster, one potential solution to explore could be utilizing spec labels to logically separate them and run them sequentially e.g. "Predictor Load", "Predictor Update", "Runtime Inference", etc.

[ckadner]

We have two separate FVT setups:

• one on Github running as part of the PR checks using minikube which is resource-constraint (.github/workflows/run-fvt.yml)
• one on IBM Cloud running a nightly e2e-pipeline that also pushes nightly builds (scripts/deploy/iks/run-perf-test.sh) which can beef up

So we could increase resources for our nightly FVTs but unless we want to link the IKS cluster for pull request checks we would need to find a strategy to work with the limited resources GitHub provides.