[docs] n-d parallelism

[stevhliu]

refactors the current Parallelism methods doc to:

• focus on practical examples of combining strategies
• adds use-case framing for each combination to make it more actionable
• removes strategies that aren't currently supported yet (PP) or are experimental (EP)

todo:

• add link to [docs] distributed training #44420 once merged
• add link to [docs] zero + sequence parallelism #44605 once merged
• add link to [docs] tp training #44613 once merged

[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.