Test vLLM env in multi-process mode

[elatt]

This repository is public. Do not put here any private DataRobot or customer's data: code, datasets, model artifacts, .etc.

Summary

• Pass runtime param in test to turning on multiple HTTP workers
• Hack uwsgi

Rationale

The GPU predictors would benefit from more workers because the underlying vLLM server can support more concurrent requests but DRUM is blocking this.