We should remove gaie components (epp). None of the llm-d well-lit paths use ms for deploying gaie. However, we still need to keep httproute. But, the problem is that there is no reference to the inference pool anymore, and the user needs to inject that as part of the httpRoute creation.

Also, what are the implications for benchmarking? https://github.com/llm-d/llm-d-benchmark