feat: Try multiple local ports for the vllm_performance actuator port-forward before failing a measurement #214
Description
Is your feature request related to a problem? Please describe.
The vllm_performance actuator uses port-forward commands when not running inside the same cluster as the vLLM deployment.
At the moment, we port-forward the vLLM deployment port to a local port starting from 10000. If the local port is not available (e.g., in use by another application) the port-forward fails and the measurement is marked as failed as well.
Instead of failing at the first unavailable local port, we should retry on a (configurable) number of subsequent ports and then fail if none are available.