v1.0.6

fix(services): cut idle CPU churn and per-request log spam (#84)

* fix(autopilot-manager): back off MAVLink reconnect when no heartbeat

With no flight controller attached, udpin:14571 never receives a heartbeat, so the message loop tore down and recreated the UDP socket every ~3s and logged a WARNING each cycle (~29k journal lines/day plus constant socket churn). Log the disconnect once per episode, probe at 1 Hz, and rebuild the link on a capped exponential backoff (1->30s) -- skipped entirely for a udpin/udp listener, which has no peer to reconnect to.

* perf(service-manager): collect unit states in one systemctl call

get_service_statuses() spawned 'systemctl is-enabled' + 'is-active' per manifested service; the UI polls /statuses at up to 1 Hz, i.e. ~22 process launches/second. Query every unit's UnitFileState/ActiveState in a single 'systemctl show' call instead (37ms vs 131ms on a Pi 5, 1 spawn vs 22). Also route the per-call print()s through a level-gated logger so the polled /statuses and /logs no longer flood the journal.

* fix(ark-ui-backend): stop double-logging every proxied request

Each request was logged twice -- morgan('dev') plus a custom console.log middleware -- with all four proxies at logLevel 'debug', making the gateway the top journal producer on an idle device. Log only 4xx/5xx via morgan, drop the redundant middleware, and lower the proxies to 'warn'. Also remove the body-logging middleware, which printed raw request bodies including WiFi PSKs.

* perf(system-manager): sample psutil memory/disk once per /info

get_common_info() called psutil.virtual_memory() and disk_usage('/') four times each per /info request; sample each once and reuse.