Hierarchical Keyring: cold-cache stampede — N concurrent decrypts → N DynamoDB/KMS calls

### Security issue notifications

If you discover a potential security issue in the AWS Encryption SDK we ask that you notify AWS Security via our [vulnerability reporting page](https://aws.amazon.com/security/vulnerability-reporting/). Please do **not** create a public GitHub issue.

### Problem:

The Node Hierarchical Keyring doesn't de-dupe concurrent branch-key lookups. If I fire a lot of decrypts for the same branch key at once against a cold cache, they all miss the cache together (it's only filled after the keystore call returns), so each one hits the keystore on its own.

So instead of one lookup I get N DynamoDB `GetItem` + N KMS `Decrypt` calls. Easy to repro: `await Promise.all` of ~3000 decrypts for the same key version, and you see ~3000 keystore calls instead of 1. Encrypt has the same problem since it shares the same code path.

### Solution:

Add single-flight to `getBranchKeyMaterials`: on a miss, the first caller starts the keystore fetch and stores the in-flight promise (keyed by cache entry id); everyone else for the same key awaits that promise instead of starting their own. The entry is dropped once it settles, so the materials cache still owns caching and TTL, and a failed request isn't shared — the next call just retries.

### Out of scope:

The legacy caching CMM has the same gap but it's a separate path, so I'm not touching it here.

[//]: # (NOTE: If you believe this might be a security issue, please email aws-security@amazon.com instead of creating a GitHub issue. For more details, see the AWS Vulnerability Reporting Guide: https://aws.amazon.com/security/vulnerability-reporting/ )

Comienza en la ruta getBranchKeyMaterials de Node Hierarchical Keyring y sigue cómo la caché de materiales gestiona una consulta con la caché fría. Reproduce el problema con descifrados o cifrados concurrentes para una versión de clave de rama y verifica que se comparta una única solicitud de keystore en curso, que se eliminen las entradas resueltas y que una solicitud fallida se pueda reintentar.

Escrito por el modelo de indexación a partir del texto del issue.