refactor(vLLM): Move video support from example to backend

[rmccorm4]

Overview:

• replace model-name allowlists with capability-driven vision loading and multimodal handling
• add native video_url loading in the standard TokensPrompt multi_modal_data flow
• move the video agg/disagg launch scripts under examples/backends/vllm and update docs/tests

Details:

Quick Benchmark: Dynamo vs vllm serve for Video Inference

I ran a quick apples-to-apples comparison between Dynamo aggregate mode (examples/backends/vllm/launch/video_agg.sh) and plain vllm serve, both serving Qwen/Qwen2-VL-2B-Instruct on the same machine and GPU configuration.

Benchmark command:

aiperf profile \
  --model Qwen/Qwen2-VL-2B-Instruct \
  --endpoint-type chat \
  --endpoint /v1/chat/completions \
  --url localhost:8000 \
  --video-width 640 \
  --video-height 480 \
  --video-fps 4 \
  --video-duration 5.0 \
  --request-count 20 \
  --osl 1200 \
  --osl-stddev 0 \
  --extra-inputs '{"ignore_eos": true, "min_tokens": 1200}' \
  --use-server-token-count \
  --ui none \
  --no-server-metrics \
  --no-gpu-telemetry

Both runs completed successfully with identical prompt/completion lengths:

• Average ISL: 962
• Average OSL: 1200
• Success rate: 20/20

Metric	Dynamo (video_agg.sh)	vllm serve	Delta
Request throughput	0.17110 req/s	0.17150 req/s	vLLM +0.23%
Avg latency	5842.53 ms	5829.02 ms	vLLM -0.23%
P50 latency	5648.22 ms	5631.58 ms	vLLM -0.30%
P90 latency	5688.17 ms	5665.62 ms	vLLM -0.40%
P99 latency	8735.89 ms	8833.25 ms	vLLM +1.11%
Output token throughput	205.32 tok/s	205.79 tok/s	vLLM +0.23%
Total token throughput	369.92 tok/s	370.77 tok/s	vLLM +0.23%
Benchmark duration	116.89 s	116.62 s	vLLM -0.23%

Where should the reviewer start?

Related Issues: (use one of the action keywords Closes / Fixes / Resolves / Relates to)

• closes GitHub issue: #xxx

[github-actions]

🌿 Fern Docs Preview: https://nvidia-preview-331dcd89-8549-4d1d-bccc-fbf9d89a5ebd.docs.buildwithfern.com/dynamo/dev