feat(scheduling): replica spread + resource-aware auto placement (#56)

services.spread distributes replicas across every online, schedulable
agent (replica i → agents[i mod n], deterministic). auto placement now
orders by real headroom — memory utilisation, cpu usage, then container
count — instead of count alone.

Lifecycle (start/stop/restart/delete) and GET /runtime are now driven
by container_instances inventory, so remote replicas are visible and
tear down wherever they run — covers spread, pins, and stale pin
leftovers. Spread and explicit pins are mutually exclusive.

Generated with [Devin](https://devin.ai)

Co-authored-by: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

app/backend/sqlc/queries/agents.sql

@@ -116,6 +116,8 @@ WHERE node_agent_id = $1 AND type = $2
 ORDER BY created_at DESC LIMIT 1;
 
 -- name: PickLeastLoadedAgent :one
+-- Resource-aware: lowest memory utilisation first, then cpu, then container
+-- count. Agents without telemetry sort last (NULL → worst score).
 SELECT a.id FROM node_agents a
 LEFT JOIN LATERAL (
     SELECT h.container_count FROM agent_heartbeats h
@@ -123,9 +125,26 @@ LEFT JOIN LATERAL (
     ORDER BY h.timestamp DESC LIMIT 1
 ) h ON true
 WHERE a.status = 'online' AND a.schedulable
-ORDER BY COALESCE(h.container_count, 0) ASC, a.created_at ASC
+ORDER BY
+    COALESCE(
+        (a.resources->'memory'->>'used')::double precision
+        / NULLIF((a.resources->'memory'->>'total')::double precision, 0),
+        1.0
+    ) ASC,
+    COALESCE((a.resources->'cpu'->>'usage')::double precision, 100.0) ASC,
+    COALESCE(h.container_count, 0) ASC,
+    a.created_at ASC
 LIMIT 1;
 
+-- name: ListSchedulableAgents :many
+SELECT id, name FROM node_agents
+WHERE status = 'online' AND schedulable
+ORDER BY id;
+
+-- name: ListServiceContainers :many
+SELECT id, name, node_agent_id, status FROM container_instances
+WHERE service_id = $1;
+
 -- name: SetAgentSchedulable :exec
 UPDATE node_agents SET schedulable = $2, updated_at = NOW() WHERE id = $1;