Add cross-platform Python launcher (launch.py) as alternative to start_all.sh

[user3868]

Motivation

start_all.sh relies on bash, which does not work on Windows natively and can fail in certain WSL configurations. This PR adds launch.py, a pure-Python alternative that works on Linux, macOS, and Windows (WSL or native) with no extra dependencies beyond the standard library.

What this PR adds

• launch.py — a single-file Python launcher requiring no additional dependencies
• Reads gateway_port and worker_base_port from config.json, consistent with start_all.sh
• Waits for model_loaded: true in each worker's /health response before starting the gateway (same logic as start_all.sh)
• Streams color-coded real-time output from each worker and the gateway to the terminal
• Saves PIDs to tmp/launch.pids for clean shutdown via --stop or Ctrl-C

Usage

# After install.sh, instead of start_all.sh:
python launch.py

# Specific GPUs:
CUDA_VISIBLE_DEVICES=0,1 python launch.py
# or
python launch.py --gpus 0,1

# HTTP mode (not recommended — browser mic/camera require HTTPS):
python launch.py --http

# Stop all services:
python launch.py --stop

CLI options

Flag	Default	Description
--gpus	CUDA_VISIBLE_DEVICES or 0	Comma-separated GPU IDs
--http	off	Use HTTP instead of HTTPS
--python	auto-detected	Path to Python interpreter
--worker-timeout	600	Seconds to wait for model load
--stop	—	Gracefully stop a running instance

Tested on

• Ubuntu 22.04 via WSL2 on Windows 10

[bokesyo]

Hello @user3868 , thank you for this PR! I think it's very helpful and easy-to-understand, and I will test it on my linux server, after that I will merge this PR. Thank you! 🤗