Add ComfyUI node integration and fix compatibility issues

- Add ComfyUI custom nodes: model loader, t2v, i2v, v2v
- Fix kernels import: move to try block for graceful fallback when not installed
- Fix diffusers compatibility: add fallback imports for HeliosLoraLoaderMixin,
  HeliosTransformer3DModel, HeliosScheduler, and apply_lora_scale
- Fix infer_helios: cast num_blocks_per_group to int, add low_cpu_mem_usage
- Add rh_config.json, model index compat, and deployment scripts

Made-with: Cursor

helios/diffusers_version/pipeline_helios_diffusers.py

@@ -25,10 +25,21 @@ from transformers import AutoTokenizer, UMT5EncoderModel
 
 from diffusers.callbacks import MultiPipelineCallbacks, PipelineCallback
 from diffusers.image_processor import PipelineImageInput
-from diffusers.loaders import HeliosLoraLoaderMixin
-from diffusers.models import AutoencoderKLWan, HeliosTransformer3DModel
+try:
+    from diffusers.loaders import HeliosLoraLoaderMixin
+except ImportError:
+    class HeliosLoraLoaderMixin:
+        pass
+from diffusers.models import AutoencoderKLWan
+try:
+    from diffusers.models import HeliosTransformer3DModel
+except ImportError:
+    from .transformer_helios_diffusers import HeliosTransformer3DModel
 from diffusers.pipelines.pipeline_utils import DiffusionPipeline
-from diffusers.schedulers import HeliosScheduler
+try:
+    from diffusers.schedulers import HeliosScheduler
+except ImportError:
+    from .scheduling_helios_diffusers import HeliosScheduler
 from diffusers.utils import is_ftfy_available, is_torch_xla_available, logging, replace_example_docstring
 from diffusers.utils.torch_utils import randn_tensor
 from diffusers.video_processor import VideoProcessor