Add S3 integration for input/output data to job executor

[dawidkurdyla]

This pull request introduces first‑class support for pulling input data from S3/MinIO and pushing output data back, while preserving backward‑compatible behaviour for existing HyperFlow jobs.

Key points:

• Data stager. New data‑stager.js and storage/s3Adapter.js modules implement:

  • Concurrent downloads from S3 prefixes/keys into the executor’s input directory. Include/exclude glob patterns and optional recursion are supported.
  • Concurrent uploads of output files back to S3, with configurable overwrite and layout (e.g. {stem}.ext to derive object names from input stems).
  • Concurrency and retries are controlled via HF_S3_CONCURRENCY and HF_S3_RETRIES (defaults sensible).
  • Optional local cleanup (HF_TASK_CLEANUP_LOCAL=1) to remove downloaded inputs and uploaded outputs after job completion.
  • Uses AWS SDK v3 (@aws-sdk/client-s3 and @aws-sdk/lib-storage) with endpoint/region/path‑style settings read from HF_S3_ENDPOINT, HF_S3_FORCE_PATH_STYLE, AWS_REGION or AWS_DEFAULT_REGION for MinIO/AWS compatibility

• Connector tweaks. The RemoteJobConnector now uses a keys object to refer to wf::tasksPendingCompletionHandling and wf::completedTasks, making it easier to mark tasks as completed or ready for completion handling

• Environment variables & defaults. New variables:

• HF_VAR_USE_S3_IO – enable S3 downloads/uploads; off by default to maintain old behaviour.

• HF_S3_ENDPOINT, HF_S3_FORCE_PATH_STYLE, AWS_REGION/AWS_DEFAULT_REGION – S3/MinIO config.

• HF_S3_CONCURRENCY, HF_S3_RETRIES – concurrency and retry controls.

• HF_TASK_CLEANUP_LOCAL – remove local data after successful upload.

• Existing workflows without these variables continue to run as before.

Dependencies. Adds @aws-sdk/client-s3, @aws-sdk/lib-storage, minimatch and updates amqplib to latest, but retains callback‑based AMQP API for backward compatibility