Maxvit model fails to run on Vulkan

🐛 Describe the bug

The maxvit_t model from torchvision fails at runtime on the Vulkan backend with "Exception raised from toTensorRef at /pytorch/executorch/../executorch/backends/vulkan/runtime/graph/containers/Value.h:266: (isTensorRef()) is false! Expected value to have type TensorRef, got TENSOR instead.".

Export repro:

import torch
import torchvision
from executorch.exir import to_edge_transform_and_lower

from executorch.backends.vulkan.partitioner.vulkan_partitioner import VulkanPartitioner

inputs = (torch.randn(1, 3, 224, 224),)
model = torchvision.models.maxvit_t().eval()

ep = torch.export.export(model, inputs)

model = to_edge_transform_and_lower(
    torch.export.export(model, inputs),
    partitioner=[VulkanPartitioner()],
).to_executorch()

It fails to run on MoltenVK for a different reason, so I don't have a nice local repro on M1 for the toTensorRef error. Logs from CI with Swiftshader can be found at https://github.com/pytorch/executorch/actions/runs/17505373586/job/49727768697.

Versions

Commit https://github.com/pytorch/executorch/commit/fbda3a9545de747329577bd910086072ec5c7ad1, M1 Mac, using MoltenVK

cc @SS-JIA @manuelcandales @cbilgin

Start by running the export repro with VulkanPartitioner and review the Vulkan backend path related to the reported toTensorRef failure. The CI Swiftshader logs linked in the issue may help compare the failure; done means maxvit_t exports and runs successfully on Vulkan without this runtime error.