Sandbox metrics lost after miren restart for recovered containers

When miren restarts and reconcileSandboxesOnBoot() recovers surviving containers, it reattaches log consumers and re-reserves IPs, but never re-registers cgroup metrics monitoring via Metrics.Add(). This means CPU and memory metrics silently stop for all recovered sandboxes and never come back.

Verified on Garden: after the most recent restart (2026-04-14 01:22:13 UTC), 13 sandboxes were successfully recovered, but only 3 sandboxes report CPU metrics (all created after the restart). The 12 surviving sandboxes have no metrics at all.

The fix should be straightforward: during reconcileSandboxesOnBoot(), after confirming a sandbox is healthy and reattaching logs, call Metrics.Add() with the sandbox's cgroup paths and log entity attributes, the same way createSandbox() does at controllers/sandbox/sandbox.go:1006.