Fix resize_mask to honor NEAREST_EXACT interpolation

[joannemiki57]

Closes #9188.

The interpolation argument was effectively ignored for tv_tensors.Mask inputs. resize_mask hardcoded InterpolationMode.NEAREST when calling resize_image, so F.resize(mask, size, interpolation=InterpolationMode.NEAREST_EXACT) produced the same output as plain NEAREST, contrary to what the docs imply.

Reproduction (before this PR):

import torch
from torchvision import tv_tensors
from torchvision.transforms.v2 import functional as F, InterpolationMode

mask = tv_tensors.Mask(torch.randint(0, 4, (1, 7, 11)))
a = F.resize(mask, size=[14, 22], interpolation=InterpolationMode.NEAREST)
b = F.resize(mask, size=[14, 22], interpolation=InterpolationMode.NEAREST_EXACT)
assert torch.equal(a, b)   # passes — bug: NEAREST_EXACT was a no-op

After this PR, b is computed with nearest-exact and the two tensors differ on inputs where the two algorithms disagree.

What changed

• Add an interpolation parameter to resize_mask and _resize_mask_dispatch, in the position expected by F.resize (mask, size, interpolation, max_size) and with the same annotation and default (Union[str, InterpolationMode, int] = InterpolationMode.BILINEAR) so test_functional_signature[F.resize_mask-Mask] passes.
• Validate through the existing _check_interpolation helper. Anything other than NEAREST / NEAREST_EXACT is silently coerced to NEAREST, preserving the historical behavior of callers that previously relied on the argument being ignored. Comparisons are chained != / and (no set literals) so the function stays TorchScript-compatible.
• Extend TestResize.test_kernel_mask to parametrize over NEAREST and NEAREST_EXACT, exercising eager / scripted / batched / CUDA via check_kernel.

A prior attempt at the same fix is open as #9384; the maintainer review there spelled out the signature ordering, set-literal, and default-value issues that this PR addresses from scratch.

Test plan

• pytest test/test_transforms_v2.py -k "TestResize and (test_kernel_mask or test_functional_signature)" -v covers the new parametrization and the signature contract (to be exercised on CI).
• Existing call sites of resize_mask (resized_crop_mask at _geometry.py:2845) pass (mask, size) positionally and are backward-compatible with the new signature.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/vision/9497

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

This comment was automatically generated by Dr. CI and updates every 15 minutes.