panic: nil pointer dereference in events recorder during reconcile

[ifdotpy]

Bug

When init-agent reconciles a LogicalCluster in the per-WorkspaceType init controller, it can panic with:

```
panic: runtime error: invalid memory address or nil pointer dereference
```

Stack trace points to `multicluster-provider/pkg/events/recorder.getBroadcaster`:

```
github.com/kcp-dev/multicluster-provider/pkg/events/recorder.(*Provider).getBroadcaster(0x0, 0xc00072a488, ...)
forked_recorder.go:105
github.com/kcp-dev/multicluster-provider/pkg/events/recorder.(*lazyRecorder).ensureRecording.func1()
forked_recorder.go:179
sync.(*Once).doSlow ...
github.com/kcp-dev/multicluster-provider/pkg/events/recorder.(*lazyRecorder).Eventf(0xc0005320a0, ...)
forked_recorder.go:196
github.com/kcp-dev/init-agent/internal/controller/initcontroller.(*Reconciler).Reconcile
reconciler.go:78
```

The receiver `*Provider` is nil at the call site (`0x0`). The lazy recorder calls `ensureRecording` and dereferences a nil broadcaster provider.

The panic is recovered by controller-runtime so reconciliation eventually succeeds on retry, but logs are noisy and the reconcile gets requeued unnecessarily.

Environment

• init-agent: built from main, commit 6ad35e2 (PRs feat: support virtual workspace URL path in --config-workspace #19 + feat: support multiple InitTargets for the same WorkspaceType #20 merged)
• multicluster-provider: v0.3.4-0.20260114155146-4b148fae0309
• Setup: --config-workspace pointed at an APIExport virtual workspace path (`/services/apiexport//initialization.kcp.io/clusters/*`)
• WorkspaceType being initialized: `account` at `root` (extends `universal`)
• Initializer name: `root:account` (set via `spec.initializer: true`)

Steps to reproduce

1. Configure init-agent --config-workspace as an APIExport virtual workspace path.
2. Create an InitTarget pointing at `workspaceTypeRef: {path: root, name: account}`.
3. Create a workspace whose effective WorkspaceType chain includes `root:account` with `spec.initializer: true`.
4. Watch init-agent logs as it processes the cluster.

Suggested fix

Either:

• Initialize `*Provider` properly before any reconcile (init-agent setup path needs to call something), or
• Guard the receiver in `getBroadcaster` so it returns a no-op recorder when no broadcaster is configured.