Real-time per-task emissions estimation for AI inference scheduling

Before deciding where a job should run, the system estimates how much carbon that job would create on each machine by combining task energy, data-center overhead, hardware health, and local grid cleanliness.