docs: add differentiability warning for Warp with mode='nearest' (#8629)

Fixes #8540 .

### Description

This PR updates the docstring of monai.networks.blocks.Warp to clarify
that mode='nearest' results in zero gradients almost everywhere,
effectively blocking gradient flow in registration training.

### Changes:

Added a note to the Warp docstring about the gradient behavior of
mode='nearest'.

Added a reference to the semi-supervised registration tutorial:

https://github.com/Project-MONAI/tutorials/blob/main/3d_registration/learn2reg_oasis_unpaired_brain_mr.ipynb

### Next Steps (future PR):

Extend VoxelMorph.forward to handle optional label and keypoint
arguments for consistent warping behavior.

### 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.
- [x] Documentation updated, tested `make html` command in the `docs/`
folder.

---------

Signed-off-by: Kheil-Z <ziadishappy@gmail.com>
Co-authored-by: Eric Kerfoot <17726042+ericspod@users.noreply.github.com>

Author: Kheil-Z

monai/networks/blocks/warp.py

@@ -52,6 +52,13 @@ def __init__(self, mode=GridSampleMode.BILINEAR.value, padding_mode=GridSamplePa
             Define reference grid on non-integer values
             Reference: B. Likar and F. Pernus. A heirarchical approach to elastic registration
             based on mutual information. Image and Vision Computing, 19:33-44, 2001.
+
+        Note that using ``mode="nearest"`` makes the warping operation effectively non-differentiable:
+        gradients are zero almost everywhere, which can block gradient flow during training.
+        For learning-based registration, use ``"bilinear"`` (2D) or ``"trilinear"`` (3D) interpolation instead.
+
+        See https://github.com/Project-MONAI/tutorials/blob/main/3d_registration/learn2reg_oasis_unpaired_brain_mr.ipynb
+        for examples of semi-supervised registration using segmentations.
         """
         super().__init__()
         # resolves _interp_mode for different methods