nx_mesh_stream.nx

nx_mesh_stream.nx -- WORKER MESH: STREAMING responses (the Triton decoupled / OpenAI-stream analog). A long job (image diffusion steps, video frames, LLM tokens) should report progress AS IT HAPPENS, not block until done. This emits a valid Server-Sent-Events stream (text/event-stream, the same shape a browser EventSource + our /api/events already speak): an `accepted` event, then one `progress` event per step (step/total/pct), then a terminal `result` event carrying the artifact URL. A client renders a live progress bar and knows exactly when the job is done. This is the streaming PROTOCOL/emitter (the sovereign contribution); the worker-side hook (real diffusion-step or frame callbacks) feeds the step numbers. For video on the west 3090 this is the natural frame-delivery channel. CLI: (no args) -> self-test GATE (SSE framing + one progress-per-step + monotonic pct + terminal result) stream <total> -> emit a real text/event-stream to stdout (the HTTP response body in a daemon) NO fake greens: the gate parses the emitted bytes back and proves the frame count, that pct rises 0->100, and that the stream TERMINATES with exactly one result (a client can detect completion). license_tier: ORIGINAL