[ET-VK] Do not partition constant_pad_nd with a symbolic pad

[msluszniak]

Summary

A pad amount derived from a dynamic dimension is serialized as a VALUELIST of Int/SymInt, and Pad.cpp reads the list with get_int_list(), which requires a literal IntList. The partitioner claims the node regardless, so the model lowers and then aborts at prepack:

Exception raised from toIntList at .../graph/containers/Value.h:272:
(isIntList()) is false! Expected value to have type IntList, got VALUELIST instead

This is reachable from ordinary code. Any model that pads a dynamic sequence up to the static length its LSTM wants hits it, which is how kokoro's synthesizer fails:

x = torch.nn.functional.pad(x, (0, 0, 0, target - seq_len))

Why not just read the list symbolically

Split.cpp, View.cpp and Expand.cpp all handle a symbolic list with extract_int_or_symint_list(), and Pad.cpp was simply missed. Swapping the extractor in is not sufficient on its own, though: add_constant_pad_nd_node() folds the amounts into a per-dim offset and bakes that into a params buffer at build time, so the dispatch would use stale offsets even if the list were read symbolically. Refreshing that buffer on resize is the real fix and is left for a follow-up. Until then this declines the node so it falls back instead of aborting.

Test plan

test_vulkan_backend_constant_pad_nd pads by a literal (1, 2, 3, 4, 5, 6), so no test covered a symbolic pad. test_vulkan_backend_constant_pad_nd_symbolic_pad aborts on main and falls back cleanly with this change.

Note on scope

This PR previously also guarded clamp/hardtanh and _native_batch_norm_legit_no_training.

• The clamp guards are dropped: [ET-VK][ops] Extend arange, clamp, and index.Tensor support #22481 implements a real dynamic-bounds path for clamp and hardtanh via add_dynamic_clamp_node, which is strictly better than declining, and keeping the guard alongside it would decline nodes that now work. hardshrink and leaky_relu still read their bounds at build time through get_val_or_inf(), but a symbolic lambda or negative_slope is a hyperparameter no real model produces, so that is left alone rather than guarded.
• The batch norm guard moved to its own PR, since it is an unrelated failure.

Merges cleanly with #22481.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/22399

• 📄 Preview Python docs built from this PR

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

⚠️ 16 Awaiting Approval

As of commit 77c1a8c with merge base 135a109 ():

AWAITING APPROVAL - The following workflows need approval before CI can run:

• Build documentation (gh)
• Build Presets (gh)
• Cadence Build & Test (gh)
• Lint (gh)
• periodic (gh)
• pull (gh)
• Test ARM Backend (gh)
• Test CoreML Backend (gh)
• Test Cortex-M Backend (gh)
• Test NXP Backend (gh)
• Test QNN Backend (gh)
• Test QNN Windows MSVC build (gh)
• Test Vulkan Backend (gh)
• Test Vulkan Backend (specialized runners) (gh)
• Test WebGPU Backend (gh)
• Test XNNPACK Backend (gh)

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.