fix(k8s): add PodDisruptionBudgets for api, ingest, and operator

## Problem
No `PodDisruptionBudget` exists anywhere in the platform. A node drain during a rolling upgrade or maintenance window can simultaneously evict all replicas of api, ingest, or the operator, causing a complete control-plane outage.

## Fix
Add PDB manifests with `minAvailable: 1` for each multi-replica service:

```yaml
apiVersion: policy/v1
kind: PodDisruptionBudget
metadata:
name: mcp-api
spec:
minAvailable: 1
selector:
matchLabels:
app: mcp-api
```

Apply the same pattern for `ingest`, `processor`, and the operator.

## Files
- `k8s/08-api.yaml`
- `k8s/06-ingest.yaml`
- `k8s/07-processor.yaml`
- `config/` (operator deployment)