Vectorized Predict function for Linear Policies

[nagireddyakshay]

Changes/Optimizations

• Vectorized the _predict_contexts() function across all Linear Policies and added a new fn _vectorized_predict_context()

Outcomes

• The revised code is 6x faster than the original implementation

Reproducibility

• Old fn creates a new random object for each context provided for reproducibility as the fn runs in parallel
• New fn don't require any of the above and utilizes only one random object for all the contexts based of a seed value
• Outcome- Reproducability retained

Changes to the codebase

• The values in the assert fn across LinTS, LinGreedy and Parallel tests were modified to accommodate changes from the new fn
• Cause -The use of single seed value across all the contexts produces a different reproducible result