Arm backend: Use FP64 conv in NSS quantized ref

[FabulousSuperDude]

Temporarily evaluate reference convolutions in FP64 for the random-data INT test, casting each result back to FP32 to reduce sensitivity to host-dependent bias accumulation. Keep calibration, export, and qtol unchanged.

Assert that the reference graph contains all 14 expected convolutions so the FP64 override cannot silently become ineffective. Preserve the existing quantization-stage settings when replacing the stage.

Validation: NSS random-data INT test and file-specific lint pass on Ubuntu.

Temporary workaround for MLETORCH-2609.

Authored with assistance from OpenAI Codex.

cc @digantdesai @freddan80 @per @zingo @oscarandersson8218 @mansnils @Sebastian-Larsson @robell @rascani

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/23259

• 📄 Preview Python docs built from this PR

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

❌ 2 New Failures, 2 Pending, 3 Unrelated Failures

As of commit 07bfe01 with merge base bfdfe10 ():

NEW FAILURES - The following jobs have failed:

• Cadence Build & Test / Resolve CI docker image / resolve (gh)
  ##[error]Refusing to check out fork pull request code from a 'pull_request_target' workflow. This workflow runs with the base repository's GITHUB_TOKEN, secrets, default-branch cache scope, and runner access. Fetching and executing a fork's code in that trusted context commonly leads to "pwn request" vulnerabilities. To opt in, review the risks at https://gh.io/securely-using-pull_request_target and set 'allow-unsafe-pr-checkout: true' on the actions/checkout step.
• pull / unittest / macos / macos-job (gh)
  RuntimeError: Command bash /Users/ec2-user/runner/_work/_temp/exec_script failed with exit code 1

BROKEN TRUNK - The following jobs failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

• pull / test-arm-backend-no-driver (test_pytest_models_tosa) / linux-job (gh) (trunk failure)
  RuntimeError: Command bash /__w/_temp/exec_script failed with exit code 1
• trunk / test-arm-backend-vkml (test_pytest_models_vkml) / linux-job (gh) (trunk failure)
  RuntimeError: Command bash /__w/_temp/exec_script failed with exit code 1
• Windows MSVC Build / build-windows-arm64 (gh) (trunk failure)

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

[FabulousSuperDude]

@pytorchbot label 'release notes: none'

[FabulousSuperDude]

@pytorchbot label 'partner: arm'

[zingo]

We have a lot of errors in this tests here on github (some/all? before this PR) IThis might add 2 new, I think it was 7 before and 9 now. Would be nice to land a fix before pile:ing up to much errors and PRs if we need to revert.

FAILED backends/arm/test/models/test_nfru.py::test_nfru_prequantized_vgf_INT[real_data] - OSError: [Errno 30] Read-only file system: '/mnt/hf_cache/hub/models--Arm--neural-frame-rate-upscaling'
FAILED backends/arm/test/models/test_nfru.py::test_nfru_vgf_quant_a16w8[real_data] - AttributeError: 'VgfPipeline' object has no attribute 'set_quantization_calibration'
FAILED backends/arm/test/models/test_nfru.py::test_nfru_vgf_quant[real_data-qat] - AttributeError: 'VgfPipeline' object has no attribute 'set_quantization_calibration'
FAILED backends/arm/test/models/test_nss.py::test_nss_vgf_INT[real_data-ptq] - AttributeError: 'VgfPipeline' object has no attribute 'set_quantization_calibration'
FAILED backends/arm/test/models/test_nss.py::test_nss_vgf_INT[real_data-qat] - AttributeError: 'VgfPipeline' object has no attribute 'set_quantization_calibration'
FAILED backends/arm/test/models/test_nfru.py::test_nfru_prequantized_vgf_INT[random_data] - OSError: [Errno 30] Read-only file system: '/mnt/hf_cache/hub/models--Arm--neural-frame-rate-upscaling'
FAILED backends/arm/test/models/test_nfru.py::test_nfru_vgf_quant[real_data-ptq] - AttributeError: 'VgfPipeline' object has no attribute 'set_quantization_calibration'
FAILED backends/arm/test/models/test_nss.py::test_nss_prequantized_vgf_INT[real_data] - OSError: [Errno 30] Read-only file system: '/mnt/hf_cache/hub/tmp5cxy_q13'
FAILED backends/arm/test/models/test_nss.py::test_nss_prequantized_vgf_INT[random_data] - OSError: [Errno 30] Read-only file system: '/mnt/hf_cache/hub/tmpt88ue_7l'
= 9 failed, 134 passed, 5 skipped, 89 warnings, 18 rerun in 692.20s (0:11:32) ==

[zingo]

Did you got the Arm tests green before merging?
I see Meta seem blocked by them alos and trying to fix it also see discussions in #23182 for more info.

[zingo]

Oh I still see you had

[pull / test-arm-backend-no-driver (test_pytest_models_tosa) / linux-job](https://hud.pytorch.org/pr/pytorch/executorch/23259#109540291065) ([gh](https://github.com/pytorch/executorch/actions/runs/36607501657/job/109540291065)) ([trunk failure](https://hud.pytorch.org/pytorch/executorch/commit/bfdfe1072fa7406ba3e5e121f89d271a5b624698#109533651389))
RuntimeError: Command bash /__w/_temp/exec_script failed with exit code 1
[trunk / test-arm-backend-vkml (test_pytest_models_vkml) / linux-job](https://hud.pytorch.org/pr/pytorch/executorch/23259#109540352941) ([gh](https://github.com/pytorch/executorch/actions/runs/36607533120/job/109540352941)) ([trunk failure](https://hud.pytorch.org/pytorch/executorch/commit/bfdfe1072fa7406ba3e5e121f89d271a5b624698#109533628759))
RuntimeError: Command bash /__w/_temp/exec_script failed with exit code 1

Bummer.

[zingo]

Meta tried to fix the problem during the night and manage to fix 2/3 problems
#23265
#23268
Left after this are merged seem to only be a few numerical/high "mean abs error" problems.