Graceful handling of cpp extensions

[msaroufim]

This fixes issues from #288

More specifically

1. We introduce a new env variable so we can install ao locally without building cpp extensions USE_CPP=0 pip install . this is useful out of convenience for faster iteration cycles
2. If the extensions were built but failed for whatever reason on a host device then importing those extensions will be skipped and ao will not crash. This also means you can import torchao from inside torchao/
3. If cpp extensions aren't build then the fp6 and ops tests are skipped

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/296

• 📄 Preview Python docs built from this PR

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

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

• Upgrade MacOS runner to 14

✅ No Failures

As of commit ac44a52 with merge base 4c1d568 ():
💚 Looks good so far! There are no failures yet. 💚

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

[msaroufim]

Failure in CI seems unrelated so review would still be helpful

[vayuda]

On wsl2:
USE_CPP=0 pip install -e. runs into https://pastebin.com/CMhYTn20
but USE_CPP=0 python setup.py develop works

[vayuda]

pip installing wheel fixed this issue. Maybe add this to dev-requirements.txt

[gau-nernst]

Looks good to me. Still not sure why some people still face problems. Without a reproducible way to trigger the problems, it is hard to see what would solve them.

[vayuda]

Not sure what the standard should be, but I know other tests (ex. quantization) import some boolean like "TORCH_VERSION_AFTER_2_4" and use that instead of a try except block

[jerryzh168]

I think the main problem is we are not sure why/when the import fails, so the boolean probably won't help here..