[float8nocompile] Add column major version of fp8 conversion kernel

[danielvegamyhre]

Summary

1. Add column major version of fp8 conversion kernel.
2. Update unit tests to validate row-major and column-major versions of kernel.
3. Refactor enabling atomic_max approach and reduction approach both use the same shared row-major/column-major conversion kernels.

Test Plan

1. pytest fp8_dynamic_tensorwise_test.py is passing

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/1465

• 📄 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 7978341 with merge base 52a5137 ():
💚 Looks good so far! There are no failures yet. 💚

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