Add nemotron-3-diarization endpoint docs and Python clients

[A-K-Erol]

Documentation and Python clients for NVIDIA Nemotron 3 Diarization on Baseten, in three presets. Docs only; the trusses are deployed from the Model Library.

Preset	Directory	Transport	What you get
Batch diarization	batch/	HTTP /predict	speaker turns for a recording; latency profile per request
Streaming diarization	streaming/	WebSocket	live turn list on every chunk; latency profile per connection
Real-time diarized transcription	diarized-transcription-streaming/	WebSocket	speaker-tagged words with timings, per-speaker partials, overlap flags

Each directory has a README covering the full request/response contract (every field, type, default, error frame, limit) and a runnable client.py (curl.sh too for batch). Clients read BASETEN_API_KEY and MODEL_ID. Example responses were captured from live deployments; test audio is public domain.

Deploy links in the READMEs resolve once the Model Library listings are published with NVIDIA's release on 2026-09-23.

🤖 Generated with Claude Code

https://claude.ai/code/session_01TtgGVtc5XtppxgMAR84vKN

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}