fix: resync Grace-Blackwell patches with current ComfyUI master

The mounted patches/model_management.py and patches/utils.py were authored
against an older ComfyUI, but COMFYUI_REF=master clones the latest. Upstream
added the DynamicVRAM/AIMDO system, and main.py now calls
model_management.get_all_torch_devices() (13 functions were missing in total),
causing comfyui to crash-loop on startup with AttributeError.

Regenerated both patches from the current master files and re-applied the
documented Sparky edits on top so they stay API-compatible:
- model_management.py: unified-memory detection, NORMAL_VRAM retention,
  95% weight ratio, intermediate_device()->cuda, soft_empty_cache skip
- utils.py: copy=False tensor load on unified memory

comfyui now starts cleanly with DynamicVRAM enabled and the Sparky
unified-memory path active.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

patches/utils.py

@@ -85,8 +85,9 @@ _TYPES = {
 def load_safetensors(ckpt):
     import comfy_aimdo.model_mmap
 
-    f = open(ckpt, "rb", buffering=0)
+    file_lock = threading.Lock()
     model_mmap = comfy_aimdo.model_mmap.ModelMMAP(ckpt)
+    f = model_mmap.get_file_handle()
     file_size = os.path.getsize(ckpt)
     mv = memoryview((ctypes.c_uint8 * file_size).from_address(model_mmap.get()))
 
@@ -111,9 +112,8 @@ def load_safetensors(ckpt):
                 storage = tensor.untyped_storage()
                 setattr(storage,
                         "_comfy_tensor_file_slice",
-                        comfy.memory_management.TensorFileSlice(f, threading.get_ident(), data_base_offset + start, end - start))
+                        comfy.memory_management.TensorFileSlice(f, file_lock, data_base_offset + start, end - start))
                 setattr(storage, "_comfy_tensor_mmap_refs", (model_mmap, mv))
-                setattr(storage, "_comfy_tensor_mmap_touched", False)
                 sd[name] = tensor
 
     return sd, header.get("__metadata__", {}),
@@ -1020,10 +1020,11 @@ def bislerp(samples, width, height):
 
 def lanczos(samples, width, height):
     #the below API is strict and expects grayscale to be squeezed
-    samples = samples.squeeze(1) if samples.shape[1] == 1 else samples.movedim(1, -1)
+    if samples.ndim == 4:
+        samples = samples.squeeze(1) if samples.shape[1] == 1 else samples.movedim(1, -1)
     images = [Image.fromarray(np.clip(255. * image.cpu().numpy(), 0, 255).astype(np.uint8)) for image in samples]
     images = [image.resize((width, height), resample=Image.Resampling.LANCZOS) for image in images]
-    images = [torch.from_numpy(np.array(image).astype(np.float32) / 255.0).movedim(-1, 0) for image in images]
+    images = [torch.from_numpy(t).movedim(-1, 0) if (t := np.array(image).astype(np.float32) / 255.0).ndim == 3 else torch.from_numpy(t) for image in images]
     result = torch.stack(images)
     return result.to(samples.device, samples.dtype)
 
@@ -1452,3 +1453,9 @@ def deepcopy_list_dict(obj, memo=None):
     memo[obj_id] = res
     return res
 
+def bit_reverse_range(index, bits):
+    result = 0
+    for _ in range(bits):
+        result = (result << 1) | (index & 1)
+        index >>= 1
+    return result