[Bug]: Critical! GraphQL health checks are consistently slower than REST checks

GraphQL entity health checks consistently take 1.5-2x longer than REST checks for the same entity, leading to false `Unhealthy` results when thresholds are set based on expected query time.

## Expected

Health checks for the same entity should have comparable response times regardless of the API surface (REST vs GraphQL), since both execute the same underlying SQL query.

## Actual

Observed across multiple test runs:

| Entity | REST (ms) | GraphQL (ms) | Ratio |
|--------|-----------|-------------|-------|
| Todo | 254-306 | 553-663 | ~2x |
| User | 255-306 | 552-663 | ~2x |
| Category | 237-306 | 570-663 | ~2x |
| Product | 306 | 663 | ~2x |

This means a `threshold-ms: 500` that comfortably passes REST checks will fail GraphQL checks on first call, making the overall health status `Unhealthy` despite the database being fine.

**Root cause:** The health check implementation makes real HTTP calls to its own REST and GraphQL endpoints. GraphQL has additional overhead (Hot Chocolate query pipeline, parsing, validation, resolver execution) compared to REST's direct controller action.

**Impact:** Users setting thresholds based on database query performance will see unexpected `Unhealthy` status. The `threshold-ms` effectively needs to account for the full middleware + framework overhead, not just DB time.

同じ threshold-ms 値を使用して、Todo、User、Category、Product エンティティのヘルスチェックを REST と GraphQL の両方で再現します。ヘルスチェック実装が REST および GraphQL エンドポイントに対して実際に行う HTTP 呼び出しを追跡し、ミドルウェアとクエリパイプラインのオーバーヘッドを比較します。API サーフェスだけが原因で、同等のチェックが誤って Unhealthy 結果を返すことがなくなれば完了です。

索引モデルが issue の本文から書いたものです。