Fix: Update s3 data type for fine-tuning jobs to support s3 uri

[mollyheamazon]

Issue #, if available:
Currently directly passing a s3 uri with jsonl file format

sft_trainer = SFTTrainer(
    ...
    training_dataset="s3://olympus-cft-dev-us-east-1/mm-cft/bedrock-input/nova-2-0/llava-cot-cleaned-300-reasoning_converse.jsonl"
)

would throw an error:

ClientError: Invalid input error: Input 'train' data file not found. Please verify that S3DataType is specified as Converse and the 'train' data channel was provided for the job, exit code: 0

Description of changes:
Update the s3_data_type from S3Prefix to Converse, since the _convert_input_data_to_channels function is only used for fine-tuning jobs.

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[mollyheamazon]

This is a feature request rather than a bug fix. Would need further discussion on the design.