Add scale benchmark tests and document performance limits

[vriesdemichael]

Summary

There is no scale testing or documented performance characteristics for the operator. Users don't know how many realms/clients the operator can handle or when they should deploy multiple operators.

Problem

Currently unknown:

• Maximum realms per operator instance
• Maximum clients per realm
• Reconciliation latency at scale (100, 500, 1000 resources)
• Memory usage growth with resource count
• CPU usage during mass reconciliation events

Proposed Solution

1. Create Benchmark Test Suite

# tests/benchmark/test_scale.py
@pytest.mark.benchmark
class TestScaleCharacteristics:
    async def test_100_realms_reconciliation_time(self):
        """Measure time to reconcile 100 realms from scratch."""
        
    async def test_500_clients_per_realm(self):
        """Measure reconciliation with 500 clients in single realm."""
        
    async def test_operator_memory_at_scale(self):
        """Track RSS memory with increasing resource counts."""

2. Run in CI (Optional, Long-Running)

Add workflow that runs weekly or on-demand:

• Spin up larger Kind cluster (multi-node)
• Create 100, 500, 1000 resources
• Measure and record metrics
• Fail if regression detected

3. Document Results

Add to documentation:

## Performance Characteristics

| Metric | Value | Notes |
|--------|-------|-------|
| Realms per operator | ~100 | Tested configuration |
| Clients per realm | ~500 | Beyond this, consider sharding |
| Reconciliation p99 | <5s | With rate limiting at defaults |
| Memory per 100 resources | ~50MB | Approximate growth |

Acceptance Criteria

• Benchmark test suite created
• Results documented in docs/operations/scaling.md
• Guidance on when to deploy multiple operators
• Memory and CPU recommendations based on scale

References

• ADR-049: Horizontal scaling via multiple operator deployments
• Review finding: TODO/SENIOR_SRE_REVIEW.md