Bug

When init-agent picks up an InitTarget pointing at a WorkspaceType it hasn't seen yet, `targetcontroller` calls `createMulticlusterManager()` which calls `mcmanager.New(wstConfig, provider, manager.Options{Scheme: scheme})` without specifying `Metrics.BindAddress`. controller-runtime defaults to `:8080`, which is already in use by the FIRST manager.

Result: the new init controller fails to start with:

```
Failed to run multicluster manager ctrlkey=root:account error=failed to start metrics server: failed to create listener: listen tcp :8080: bind: address already in use
```

The new InitTarget is effectively ignored until the pod restarts (and it happens to start with that target already known).

Reproduction

1. Run init-agent with one InitTarget A targeting WT `foo`.
2. Apply a second InitTarget B targeting WT `bar`.
3. init-agent's targetcontroller logs "Creating new init controller…" and then crashes the new manager on port collision.

Suggested fix

In `internal/controller/targetcontroller/controller.go` `createMulticlusterManager`, set `manager.Options{ ..., Metrics: metricsserver.Options{BindAddress: "0"} }` for all subsequent managers, OR allocate distinct ports automatically. Only the primary manager needs metrics.

Environment

• init-agent: main, commit 6ad35e2