Feature request: video content blocks (video/*) alongside ImageBlock - pass video to video-capable models

[Nonexistent01]

Context

DSH's typed model-content contract is image-only:

• @deepseek-ai/dsh-llm lib/typert.host.js L329-333: ImageBlock { type: 'image' } and ImageMediaType = 'image/png' | 'image/jpeg' | 'image/webp' | 'image/gif'
• A search across all 289 @deepseek-ai/* packages finds no real video/ handling anywhere (the only hit is inside the vendored pdf.js bundle string — a false positive).

Problem

Most current frontier models accept video input (via native video parts or frame sequences, over the same OpenAI-compatible / provider chat APIs that DSH already speaks). Because the framework's content model has no video block type, there is no way to hand a video clip to a video-capable model — e.g. an MCP tool returning an mp4, or a user attaching a short screen recording. The capability is lost at the framework layer, not the model layer.

Request

Add a VideoBlock { type: 'video' } analogous to ImageBlock (attachment ref + video/* media type) and plumb it end to end:

1. attachment/input acceptance (video/mp4, video/webm, …)
2. typed content + compaction/offloading rules
3. provider adapters: pass-through where the model supports video parts; graceful degradation otherwise (keyframe extraction to N images, or a text placeholder)

An MVP limited to pass-through on supporting adapters + a documented fallback would already unlock this.