Docs: update inference guide (visdet APIs + FiftyOne batching)

[GeorgePearse]

Summary

• Update the User Guide inference page to use visdet APIs (remove mmdet/mmcv/mmengine examples).
• Improve the FiftyOne/Voxel51 example to run streaming batched inference via inference_detector(model, batch_of_filepaths) and attach predictions per-batch.
• Export async_inference_detector from visdet.apis so the guide can import it directly.

Notes

• This PR is docs-focused; multi-GPU inference support is already on main.

[github-actions]

🤖 Multi-Model Consensus Review

Model	Status
Claude Sonnet 4	✅
GPT-4o	✅
Gemini 2.0 Flash	✅

✅ visdet/apis/__init__.py

Pass Rate: 3/3 models

[github-actions]

✅ Skylos Scan: No dead code or security issues detected.