Separate deployment of front-end and back-end

### Feature Idea

Problem: One major challenge in using ComfyUI is the issue of resource allocation.

Background: We have a large GPU server cluster and use Kubernetes for container scheduling. We provide each UI colleague with a ComfyUI interface and allow users to upload models and workflows.

1. Repeatedly uploaded models lead to high storage pressure and increased costs.

2. If a colleague's requested comfyui node is rarely used and they are unsure whether to continue using it, it will result in the GPU being idle and the utilization rate failing to reach the expected level. This in turn leads to an increase in costs.

Desired: It is hoped that the front-end page and the algorithm logic for handling front-end requests (including the programs that call computing power) can be separated. This way, the model files can be managed uniformly. If colleagues A and B both apply for comfyui, I only need to deploy one front-end container for each of them. If the computing power is insufficient, I can increase the number of replicas of the back-end computing power. If other colleagues have other model requirements, I can upload the models they need. This achieves centralized management and greatly improves the utilization rate of computing power nodes.

[ Front-end page & workflow & custom nodes ] --> [ Back-end computing power ]

### Existing Solutions

None

### Other

We all agree that ComfyUI is an extremely perfect project. As time goes by and with the advancement of technology, it will become even more perfect.