MPDynamic: force load flux img_in weight (Fixes flux1 canny+depth lora crash) (#12446)

* lora: add weight shape calculations.

This lets the loader know if a lora will change the shape of a weight
so it can take appropriate action.

* MPDynamic: force load flux img_in weight

This weight is a bit special, in that the lora changes its geometry.
This is rather unique, not handled by existing estimate and doesn't
work for either offloading or dynamic_vram.

Fix for dynamic_vram as a special case. Ideally we can fully precalculate
these lora geometry changes at load time, but just get these models
working first.

comfy/lora.py

@@ -374,6 +374,31 @@ def pad_tensor_to_shape(tensor: torch.Tensor, new_shape: list[int]) -> torch.Ten
 
     return padded_tensor
 
+def calculate_shape(patches, weight, key, original_weights=None):
+    current_shape = weight.shape
+
+    for p in patches:
+        v = p[1]
+        offset = p[3]
+
+        # Offsets restore the old shape; lists force a diff without metadata
+        if offset is not None or isinstance(v, list):
+            continue
+
+        if isinstance(v, weight_adapter.WeightAdapterBase):
+            adapter_shape = v.calculate_shape(key)
+            if adapter_shape is not None:
+                current_shape = adapter_shape
+            continue
+
+        # Standard diff logic with padding
+        if len(v) == 2:
+            patch_type, patch_data = v[0], v[1]
+            if patch_type == "diff" and len(patch_data) > 1 and patch_data[1]['pad_weight']:
+                current_shape = patch_data[0].shape
+
+    return current_shape
+
 def calculate_weight(patches, weight, key, intermediate_dtype=torch.float32, original_weights=None):
     for p in patches:
         strength = p[0]