Metal backend: Add SDPA head_dim=256 support

[manuelcandales]

Qwen 3.5 MoE uses head_dim=256 for full attention layers. The existing
SDPA Metal kernel only instantiated head_dim 64, 96, 128. At D=256 each
thread handles 8 QK elements (8 x 32 threads = 256 dims); register
pressure and threadgroup memory are well within Apple GPU limits.

Authored with Claude.

[manuelcandales]

Stack from ghstack (oldest at bottom):

• Qwen 3.5 MoE Metal: Use max-sized prefill example for dynamic inputs #18956
• Qwen 3.5 MoE: Add Metal build target #18881
• Qwen 3.5 MoE: Add --backend metal export path #18880
• Qwen 3.5 MoE: Add Metal source transformations #18879
• Metal backend: Add gated delta rule kernel for linear attention #18878
• Metal backend: Add gather_qmv kernel for MoE expert-indexed quantized matmul #18877
• Metal backend: Add topk fallback kernel via MPSGraph #18876
• -> Metal backend: Add SDPA head_dim=256 support #18875
• Metal backend: Add INT32 and BOOL dtype support #18874

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/18875

• 📄 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.

[github-actions]

This PR needs a release notes: label

If your change should be included in the release notes (i.e. would users of this library care about this change?), please use a label starting with release notes:. This helps us keep track and include your important work in the next release notes.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "release notes: none"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.