H3 Studio remote API node

Extract this folder into ComfyUI/custom_nodes/comfyui_h3_studio.
Set H3_STUDIO_API_KEY in the environment that starts the ComfyUI server.
Use your studio API key, not the browser login password. Never save the key
in workflow JSON or node widgets. Restart the ComfyUI server after setting it.

Linux/macOS:
  export H3_STUDIO_API_KEY='your-studio-api-key'
  python main.py

Windows PowerShell (start ComfyUI from this same shell):
  $env:H3_STUDIO_API_KEY = 'your-studio-api-key'
  .\run_nvidia_gpu.bat

For a desktop app, Docker container, or service, set the variable in that
application's launcher/container/service environment, then restart it.

Import workflow.json in ComfyUI. Set api_base to your studio origin, e.g.
http://192.168.20.105:8088, without /v1 or /api. Connect the IMAGE output to
Save Image, Preview Image, or another image-processing node.

The node submits an asynchronous HTTP job, polls for completion, downloads
the PNG, and returns a ComfyUI IMAGE tensor [1,H,W,3] with values in [0,1].
It loads no MiniMax weights and needs no additional packages beyond the
dependencies included with a normal ComfyUI installation.

ComfyUI caches identical inputs. Change the seed or prompt for another result.
Canceling in ComfyUI stops polling; an already submitted remote job continues.
Do not immediately retry a network failure: inspect the studio's image library
first, since an accepted job may still complete on the server.

Reference: https://docs.comfy.org/custom-nodes/backend/server_overview
Image format: https://docs.comfy.org/custom-nodes/backend/datatypes
