add flux-klein example top of trickle protocol

[zargarzadehm]

Add flux-klein — realtime FLUX.2-klein-4B video-to-video over trickle

Adds a new example implementing the request in #1: realtime FLUX.2 [klein] img2img
over trickle on the live runner, modelled on the existing trickle/static-runner
examples.

App id	livepeer-example/flux-klein
Registration	static (runners.json + /health poll, like vllm)
Mode	persistent
Transport	trickle (video in/out + /update control)
Port	8720
Requires	NVIDIA GPU, ~16 GB VRAM

What it does

Receives a live video stream, transforms every frame against a text prompt with
FLUX.2-klein-4B, and streams the result back. Prompt, noise seed, and camera
anchoring are all adjustable mid-stream over /update.

Self-contained inference. flux_klein.py wraps diffusers
(Flux2KleinPipeline) directly — it does not run Daydream Scope or any plugin
runtime. The reference implementation cited in #1,
hthillman/scope-flux-klein, is a
Scope plugin; this example borrows its recipe (the Krea-style feedback loop and
the truncated-schedule refine) and reimplements it standalone so the example stays
self-contained like the others.

The feedback loop. Klein is a 4-step model, so full inference per frame is far
too slow for video. Frame 0 runs full inference; every frame after partially
denoises the previous output blended with the incoming frame. Because
Flux2KleinPipeline has no strength parameter (it conditions on the image via
joint attention), the refine hand-rolls a truncated flow-matching schedule.

Measured throughput

NVIDIA RTX 6000 Pro, 384×384, 4 steps, feedback 0.5:

Config	fps
defaults	~19
FLUX_BATCH=2 FLUX_COMPILE=max-autotune	~22.7

GET /stats reports live input/inference/output rates, frames_skipped, and the
active tuning parameters.

Files

flux_klein.py (model + feedback loop) · runner.py (aiohttp app, trickle
channels, worker) · client.py (file or webcam in, live prompt/seed/blend updates)
· Dockerfile · docker-compose.yml + docker-compose.onchain.yml ·
runners.json · pyproject.toml · .env.example · view.sh · README.md