docs: Add Wan 2.2 text2vid training documentation to README

[Toshi-31]

Description

Following the merge of PR #470 (Wan 2.2 dual-expert training pipeline), this PR updates README.md to document Wan 2.2 text-to-video training support.

Changes

1. What's new?: Added announcement for Wan 2.2 dual-expert training support.
2. Overview: Updated supported features list to include Wan 2.2 text2vid training.
3. Table of Contents: Added Wan 2.2 to the Training section.
4. Training Documentation: Added a dedicated ## Wan 2.2 Training section covering:
   • Dual-expert DiT architecture overview (~27B total parameters) with joint dynamic routing based on boundary_ratio.
   • Single-host training command using train_wan.py and base_wan_27b.yml.
   • Multi-host distributed training command via XPK.
   • Dual-expert checkpointing and Diffusers export via WanCheckpointer2_2.

[gemini-code-assist]

Code Review

This pull request updates the README.md to document support for Wan2.2 text2vid dual-expert training, including single-host and multi-host training commands. However, the feedback points out a discrepancy where the documented train_mode parameter is not actually defined in the configuration file or implemented in the trainer, which would lead to configuration parsing errors if executed.