[Op] Add attention and bias_gelu ops

[comaniac]

Description

This is a prerequisite PR for adding HF GPT-2 schedule.

• Implement attention ops that use flash-attention and xformrs.
• Implement bias_gelu ops that use torchscript or torch compiler.
• Change the GPT-Neo schedule to use these ops, so that now GPT-Neo schedule doesn't depend on epoi anymore. Later we will update the schedules of other example models accordingly.
• [Test] Add -rxXs to let pytest print reasons of skipped tests.
• [Docker] Update flash-attention commit hash which improves the kernel performance by ~12%. CI image is not updated because this change doesn't impact the functionality.

Checklist

• PR's title starts with a category (e.g. [Bugfix], [Model], [Tutorial], etc)
• Changes are complete (i.e. I finished coding on this PR)
• All changes have test coverage
• Code is well-documented

@szhengac @chhzh123

[comaniac]

Thanks @chhzh123