[Relay][Topi] Hook up LayerNorm to new Topi

[AndrewZhaoLuo]

This makes layer_norm relay op dispatch to new topi committed in PR #12864.

It removes LayerNorm from simplify inference pass now that we actually have compute.

Using TIR CSE elimination pass with FP16 layernorm also necessitates the handling of FP16 type when packing args for CUDA.

[vinx13]

I'm wondering removing simplification for later norm affects non-metaschedule workflow. We didn't implement topi schedule for layer norm, I'm not sure if the default schedule_injective works well.

Btw there are also some unrelated (.csv, .md) files in this PR added by mistake.

[AndrewZhaoLuo]

I have removed the unneeded files.

Good point about non-metaschedule workflows. I will see what I can do here.

[masahi]

Any update? Also interested in using fused layer norm.

[AndrewZhaoLuo]

@masahi

Ah yes, I will try to get to this during the next week where I will have time.

1. Figure out the behavior for autotvm and autoscheduler.
2. The topi kernel for fused layernorm is known to be numerically unstable. We should replace it with a more stable version e.g. https://en.wikipedia.org/wiki/Algorithms_for_calculating_variance#Parallel_algorithm first.
3. [Packed Func] Pack args fp16 support #13532 this PR is included here to support fp16 layernorm, but needs some refactoring to avoid casting cost. Now fixed.