Support reading S3 object meta data only

[Zhangxunmt]

Is your feature request related to a problem? Please describe.
The feature request is related to the offline batch ingestion integration with ml-commons. opensearch-project/ml-commons#2891

The S3-Source Scan currently reads the full content of S3 objects. However, for offline ML batch job processing, only object metadata (such as name and size) is required. Reading the entire content can lead to duplicate batch job executions, adding unnecessary overhead.

Describe the solution you'd like
Enhance the S3-Source Scan by introducing an option to retrieve only metadata (e.g., object name, size) instead of reading full object contents. This ensures that batch job processors receive only the required information, preventing redundant job triggers.

Additional context
This improvement would optimize offline batch ingestion while also benefiting future async processors that require metadata without processing full object content.

[dlvenable]

This sounds like a nice feature to have the S3 scan.

One question we should discuss is whether or not to put the data into the s3 key for consistency or not.

The current S3 scan will create events like:

{
  s3: {
    "key" : "my-key.txt"
    "bucket": "my-bucket"
  }
}

Should we output this? Or move these to the top level?

[Zhangxunmt]

The first version of this enhancement outputs S3 metadata like this.
{"bucket":"offlinebatch","length":6234,"time":1738108982.000000000,"key":"sagemaker/sagemaker_djl_batch_input_1.json"}