feat: add speech-to-video and animation pipelines

[ehsanrs2]

Summary

• add WanS2V and WanAnimate pipelines with configs, modules, and CLI wiring
• introduce shared text-conditioning/memory-diagnostic utilities plus CosyVoice download helpers
• document new workflows and optional dependency installs alongside unit tests

Motivation

• support speech-driven and pose-driven generation without external patches
• improve VRAM management via reusable embedding/offload helpers
• provide user-facing docs/tests for confident adoption

Implementation Notes

• CLI now accepts audio/TTS/animation flags and applies safer defaults for 5B jobs
• CosyVoice helpers respect WAN_COSYVOICE_* env overrides and guard against missing git/huggingface-hub
• Tests import modules dynamically to avoid optional dependency noise
• isort/black executed per-file due to repo size; pytest warns about CUDA on CPU sandbox

Breaking Changes

• none
  Testing

• python -m pytest tests/test_utils.py # passes (CUDA availability warning expected)