Add anomaly transformer method

[sarahmish]

Transformers are popular models for a variety of tasks including anomaly detection. There are many methods that have been developed including Anomaly Transformer that use attention for anomaly detection.

As part of our endeavor to provide popular models in Orion, we seek to provide anomaly transformer pipeline, paper reference: https://arxiv.org/pdf/2110.02642.pdf

Design
Initial plan for primitives to use

• time segments aggregate primitive to make the samples equi-distant
• imputing missing values
• scaling the data using Standard Scaler
• transformer model
• find the threshold using attention values

Evaluation Criteria
One clear difference between Orion benchmark's evaluation calculation and the paper's is that in Orion we only factor in the positive class (i.e. anomalies). In other words, we do not care if we predict normal sequences and normal. We mainly care about (1) getting all the correct anomalies (2) not raising false alarms