fix: use register_buffer for kernel and kernel_vol in LocalNormalizedCrossCorrelationLoss (#8818)

### Description

This PR fixes an incorrect kernel initialization pattern in
`LocalNormalizedCrossCorrelationLoss` introduced in
`monai/losses/image_dissimilarity.py`.

**Bug found:** The original code used plain attribute assignment with a
typo:
```python
self.kernel = _kernel(self.kernel_size)
self.kernel.require_grads = False  # typo: 'require_grads' is NOT a valid tensor attribute
self.kernel_vol = self.get_kernel_vol()
```

`require_grads` (with an 's') is not a valid PyTorch tensor attribute.
The correct attribute is `requires_grad`. This means the kernel was
**silently tracking gradients** throughout every forward pass, wasting
memory and computation.

Furthermore, assigning the kernel as a plain attribute (`self.kernel =
...`) rather than registering it as a buffer means:
- The kernel tensor does **not** automatically move to the correct
device when `.to(device)` / `.cuda()` / `.half()` is called on the loss
module
- The kernel is **not saved/restored** correctly via `state_dict()` /
`load_state_dict()`

**Fix:** Replace both assignments with `register_buffer`, which is the
correct PyTorch `nn.Module` pattern for constant tensors:
```python
self.register_buffer("kernel", _kernel(self.kernel_size))
self.register_buffer("kernel_vol", self.get_kernel_vol())
```

This removes the need for any manual `requires_grad` management and
ensures proper device placement.

### Types of changes
<!--- Put an `x` in all the boxes that apply, and remove the not
applicable items -->
- [x] Non-breaking change (fix or new feature that would not break
existing functionality).
- [ ] Breaking change (fix or new feature that would cause existing
functionality to change).
- [ ] New tests added to cover the changes.
- [ ] Integration tests passed locally by running `./runtests.sh -f -u
--net --coverage`.
- [ ] Quick tests passed locally by running `./runtests.sh --quick
--unittests --disttests`.
- [x] In-line docstrings updated.
- [ ] Documentation updated, tested `make html` command in the `docs/`
folder.

Signed-off-by: Zeesejo <92383127+Zeesejo@users.noreply.github.com>

---------

Signed-off-by: Zeeshan Modi <92383127+Zeesejo@users.noreply.github.com>
Signed-off-by: Eric Kerfoot <17726042+ericspod@users.noreply.github.com>
Co-authored-by: Eric Kerfoot <17726042+ericspod@users.noreply.github.com>

Author: Zeesejo

monai/losses/image_dissimilarity.py

@@ -111,14 +111,16 @@ def __init__(
             raise ValueError(f"kernel_size must be odd, got {self.kernel_size}")
 
         _kernel = look_up_option(kernel_type, kernel_dict)
-        self.kernel = _kernel(self.kernel_size)
-        self.kernel.require_grads = False
-        self.kernel_vol = self.get_kernel_vol()
+        self.kernel: torch.Tensor
+        self.kernel_vol: torch.Tensor
+        self.register_buffer("kernel", _kernel(self.kernel_size), persistent=False)
+        self.register_buffer("kernel_vol", self.get_kernel_vol(), persistent=False)
 
         self.smooth_nr = float(smooth_nr)
         self.smooth_dr = float(smooth_dr)
 
-    def get_kernel_vol(self):
+    def get_kernel_vol(self) -> torch.Tensor:
+        assert self.kernel is not None
         vol = self.kernel
         for _ in range(self.ndim - 1):
             vol = torch.matmul(vol.unsqueeze(-1), self.kernel.unsqueeze(0))
@@ -138,6 +140,8 @@ def forward(self, pred: torch.Tensor, target: torch.Tensor) -> torch.Tensor:
             raise ValueError(f"ground truth has differing shape ({target.shape}) from pred ({pred.shape})")
 
         t2, p2, tp = target * target, pred * pred, target * pred
+        assert self.kernel is not None
+        assert self.kernel_vol is not None
         kernel, kernel_vol = self.kernel.to(pred), self.kernel_vol.to(pred)
         kernels = [kernel] * self.ndim
         # sum over kernel