[float8nocompile]: Wrap fp8 conversion kernels in autograd func and use in Float8NoCompileLinear

[danielvegamyhre]

Summary

• Wrap the float8 compile triton kernels in an autograd func to make the conversion differentiable (for use in training)
• Integrate these autograd functions into the Float8NoCompileLinear class

Test plan

• The unit tests enforcing fidelity between float8 training path and this prototype's path are passing.

pytest test/test.py

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/1452

• 📄 Preview Python docs built from this PR

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

✅ No Failures

As of commit b8e1c6a with merge base 692236a ():
💚 Looks good so far! There are no failures yet. 💚

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

[danielvegamyhre]

adding @drisspg @vkuzo for review

[vkuzo]

nit: remove e5m2 from docstring, since the target dtype is passed as an argument

[vkuzo]

remove e5m2 from comment

[vkuzo]

looks good, please ensure CI is green before landing