[Bug]: Bad connection string health check shows "threshold exceeded" instead of connection error

When the data source connection string points to a non-existent server, the health check reports `"The threshold for executing the request has exceeded."` instead of a connection error.

Operators monitoring health endpoints cannot determine the root cause of an Unhealthy status. A threshold exceeded vs. connection failure require very different remediation actions.

## Expected

The health check should report a connection-related exception, e.g.:
```json
{
"status": "Unhealthy",
"name": "primary-sql-db",
"exception": "A network-related or instance-specific error occurred while establishing a connection to SQL Server.",
"tags": ["data-source"]
}
```

## Actual

```json
{
"status": "Unhealthy",
"name": "primary-sql-db",
"exception": "The threshold for executing the request has exceeded.",
"tags": ["data-source"],
"data": {"response-ms": 137, "threshold-ms": 1500}
}
```

The actual connection failure gets swallowed and replaced with a generic threshold message. This makes it impossible to distinguish between "database is slow" and "database is unreachable" from the health response alone.

まず、データソースのヘルスチェックのエントリーポイントと、失敗した SQL 接続をしきい値メッセージに変換する処理を特定します。次に、存在しないサーバーのケースを再現し、その後、ヘルスレスポンスが接続関連の例外を保持する一方で、実際に遅いリクエストについてはしきい値データが残ることを確認します。

索引モデルが issue の本文から書いたものです。