Polish Echo-Memory Wan video integration

[SJTUGSC]

This PR polishes the Echo-Memory Wan video integration based on PR #1592.

Changes:

• Move Echo-Memory loading into diffsynth.pipelines.
• Update standard and low-vram inference examples.
• Add LoRA/full training scripts and validation examples.
• Update README and Wan model detail tables.
• Remove the standalone examples/wanvideo/echo_memory helper folder.

Validation:

• Echo-Memory standard inference runs and saves video.
• Echo-Memory low-vram inference runs and saves video.
• LoRA training starts successfully.
• Full training starts successfully.
• py_compile, bash -n, and git diff --check passed.