Last Batch Policy changes for file source reader

[swetha097]

• Modify the way sharding is handled
• Handle the various policies , pad_last_batch, stick_to_shard and shard_size.
• A new vector is introduced to store the file_names of the complete dataset instead of just the current shard data to support stick_to_shard.
• Pass the last batch policy and pad_last_batch_repeated to the loaders instead of the context creation.
• Introduce a struct ShardingInfo to hold variables related to sharding - last_batch_policy, pad_last_batch_repeated, stick_to_shard and shard_size.

[swetha097]

@rrawther Resolved all the PR comments and tested with the Audio and Classification trainings.