DistPartition should move the raw id mappings to the partition output

[thvasilo]

Right now for the distributed pipeline we would first create raw id mappings using GSProcessing, which the second step of DistPart does not touch.

When a user runs a training job on the DistPart output, they will not be able to get embeddings with their original ids by default.

To fix this, our local and SageMaker DistPart runners, that both expect GSProcessing-produced files as input, should also copy the raw_id_mappings under the same output directory as the dist_graph is created.

Our current code assumes the raw id mappings are placed at the same level as the dist graph: https://github.com/awslabs/graphstorm/blob/7e2cdfdc91266f50f31e7d192d4a51e84dd55e55/python/graphstorm/gconstruct/remap_result.py#L655

We can also provide the option to provide the raw id mapping path during training.