parallelize ExecuTorch build

[swolchok]

No description provided.

[swolchok]

Stack from ghstack (oldest at bottom):

• export.py: fix custom SDPA type conversion logic & re-enable for bfloat16 #1193
• Default max_seq_length to 128 for ExecuTorch export #1192
• Also use default max_seq_length of 128 for ExecuTorch native runner #1191
• Use default max_seq_length of 128 when loading ExecuTorch models #1190
• add ConvertToLinear, disable custom SDPA for bfloat16 #1189
• -> parallelize ExecuTorch build #1188
• Update ExecuTorch pin to pick up bfloat16 fixes #1187

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/torchchat/1188

• 📄 Preview Python docs built from this PR

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

❌ 1 New Failure

As of commit 7808b9a with merge base 04ea309 ():

NEW FAILURE - The following job has failed:

• pull / test-mps / macos-job (gh)
  RuntimeError: Command bash /Users/ec2-user/runner/_work/_temp/exec_script failed with exit code 1

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