Configure autostop/autostart for machines to enable idle-driven lifecycle management

Documented steps

1. Add auto_stop_machines setting in [http_service] or [[services]] section of fly.toml with value 'off', 'stop', or 'suspend'
2. Set auto_stop_machines='stop' to enable Fly Proxy to halt machines during idle periods; use 'suspend' for faster recovery (if supported)
3. Add auto_start_machines=true to enable proxy to restart stopped/suspended machines when requests arrive and capacity is exceeded
4. Configure min_machines_running=<N> to specify minimum machines always running in primary region when auto-stop is active (default 0)
5. Set concurrency soft_limit per machine to define when excess capacity exists (proxy calculates: excess_capacity = num_machines - (num_over_soft_limit + 1))
6. Fly Proxy runs idle detection every few minutes and stops at most one machine per region per cycle based on soft_limit
7. For single machine: proxy checks traffic level and stops if load=0; for multiple machines: uses soft_limit calculation
8. Enable both auto_stop_machines and auto_start_machines together or disable both to prevent stuck states (machines never stopping or never restarting)
9. Autostop/autostart only manages existing machines; it never creates or destroys machines automatically
10. Deploy fly.toml changes with flyctl deploy to activate these settings
11. Docs: https://fly.io/docs/launch/autostop-autostart/