[docs] Optimize and scale pt. 2

[stevhliu]

Finishes updating the Optimize and scale section in #14867:

• adds a table to "Reduce memory usage" to help users select a strategy based on their use case
• fix the docs to match the code (default shard size is 10GB, skip_modules_pattern in the layerwise casting example, removed a loaded text encoder in Wan example that was never used, etc.)

[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.

[sayakpaul]

Will review after #14867 is merged to keep things manageable.