Avoid rereading CUDA weights blob for SHA-256 computation

[Gasoonjia]

Summary

• Compute SHA-256 while streaming the CUDA AOTI external weights blob to disk, avoiding a second full read of the approximately 19 GB weights file during serialization.
• Preserve lazy SHA-256 computation for callers that do not provide a digest.

Performance

In an independent MG 30B cold export with an empty Inductor cache, export time improved from 21:16.45 to 18:56.75, a 10.9% reduction.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21800

• 📄 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

As of commit 6b88a80 with merge base e60faa2 ():

NEW FAILURES - The following jobs have failed:

• Test CUDA Builds / export-model-cuda-artifact (SocialLocalMobile, Qwen3.5-35B-A3B-HQQ-INT4, quantized-int4-tile-packed) / linux-job (gh)
  RuntimeError: Command bash /__w/_temp/exec_script failed with exit code 1
• Test CUDA Builds / unittest-cuda-runtime / linux-job (gh)
  RuntimeError: Command docker exec -t 2dca154ce0e1de06fb393e10c49a263a866517b0b73002700a8bfe5e55e0f658 /exec failed with exit code 2

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