update pd disaggregation templates and example

[elieserr]

This include various changes

• use ghcr.io/llm-d/llm-d-cuda:v0.3.0 for prefill and decode deployments
• use ghcr.io/llm-d/llm-d-inference-scheduler:v0.3.1
• update the template for inference scheduler to match v0.3.1 args
• use default-pd-config.yaml for EPP config
• set VLLM_NIXL_SIDE_CHANNEL_PORT variable to default vLLM port
• disable default deprecated inferencemodels
• replaces rbac of inferencemodels with inferenceobjectives
• reduce cpu requirements since the model of the example fits on smaller node

Fixes #130

[kalantar]

we are thinking that the endpoint picker/inferencepool pieces should all be removed from the modelservice chart. There is an upstream chart defined here: https://github.com/kubernetes-sigs/gateway-api-inference-extension/tree/main/config/charts/inferencepool (released versions at oci://registry.k8s.io/gateway-api-inference-extension/charts/inferencepool) that has all these updates already. Can you try this and see where you are stuck?

See #135.