[docs] distributed training

[stevhliu]

Updates the distributed training docs with recent changes to the DistributedConfig API.

• Replace the deprecated tp_plan and tp_size in from_pretrained with distributed_config=DistributedConfig(tp_size=N)
• Add new DistributedConfig doc to cover TP, FSDP2, sharding plans, save_pretrained(distributed_checkpoint=True), and optimizer save/load
• Add new overview doc for how to choose a strategy and provide a better entry point to each individual parallelism strategy. Also differentiates between using Trainer/Accelerate and load-time DistributedConfig
• Extends debugging doc with other issues a user may potentially run into (taken from source) instead of mostly DeepSpeed
• Minor autolink formatting (use # instead of .)

Many of the changed files are due to formatting. The main ones to focus on are:

• distributed_config.md
• distributed_overview.md
• perf_train_gpu_many.md

[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.

[github-actions]

---

CI recap

Dashboard: View test results in Grafana
Latest run: 29037662996:1
Result: success | Jobs: 2 | Tests: 24 | Failures: 0 | Duration: 1m 57s