Adjust the dataset.py with different input settings

[cshjin]

• unify the data folder structure
• unify the anomaly cat and level setting if available
• unify the node/graph data process through args
  • all data reading goes to node-level. For graph-level, extract data by reshaping the xs and ys.
• unify the processed_data locations.
• unify dataset.py and dataset_v2.py merge dataset_v2.py and dataset.py into a single file #11
• (optional) apply pre_tranasform for normalization