[docs] Distilled inference

[stevhliu]

This PR combines "Distilled Stable Diffusion inference" with the "Speed up inference" doc. It also:

• updates the table with only "speed" results (adds tf32 and combined results) and not the "memory" related ones like channels last or traced UNet (these are kept in the "Reduce memory usage" doc)
• I don't have a Titan RTX so my results were obtained from a Colab A100
• removes the code snippets for timing the inference run in favor of keeping things simpler and allowing users to just copy and use the code

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.