feat: manage video text encoders in Media Models

[atomantic]

Summary

• Expose video text-encoder choices, including MiniMax H3 substitutes, through the Media Models catalog API.
• Add explicit download progress and confirmed cache deletion for each separately downloaded encoder.
• Keep encoder cache paths aligned with Hugging Face cache overrides and out of the orphaned cache list.

Test plan

• npm --prefix client test -- --run src/pages/MediaModels.test.jsx
• npm --prefix server test -- --run routes/imageVideoModels.test.js
• npm --prefix client run build