feat: add lucy-vton-3 (realtime + queue/batch)

[AdirAmsalem]

What

Adds lucy-vton-3 as a public virtual try-on model, available on both realtime streaming and the queue/batch API. It joins the existing VTON lineup (lucy-2.1-vton, lucy-vton-2), which all remain supported. lucy-vton-latest continues to resolve server-side to the newest VTON.

Why

Gives users access to the latest virtual try-on generation through the Python SDK, with the same input surface and ergonomics as the previous VTON models — no migration needed to adopt it.

Usage

from decart import models

# Realtime streaming
rt = models.realtime("lucy-vton-3")

# Queue / batch video try-on
result = await client.queue.submit_and_poll(
    model=models.video("lucy-vton-3"),
    prompt="Wearing a red leather jacket",
    data=video_blob,
)

---

Note

Low Risk
Additive model registry and test coverage only; no auth, billing, or behavioral changes to existing models.

Overview
Registers lucy-vton-3 in the Python SDK model catalog so callers can select it via models.realtime("lucy-vton-3") and models.video("lucy-vton-3"), alongside the existing VTON models.

Realtime maps to /v1/stream at 30 fps (1088×624), matching other VTON stream models. Queue/batch uses /v1/jobs/lucy-vton-3 at 20 fps with the same VideoEdit2Input schema as lucy-vton-2. Type literals RealTimeModels and VideoModels are updated accordingly, and tests assert registry paths and dimensions for lucy-vton-2 and the new model.

^{Reviewed by Cursor Bugbot for commit 9a02b98. Bugbot is set up for automated code reviews on this repo. Configure here.}