Batched strategies

[vreshniak]

This pull request introduces support of batch strategies.

Key changes include:

Strategy Handling and API Enhancements

• Supporting new batch strategies ('liar', 'believer'). The strategy argument types were also expanded to allow lists for more flexible parameterization.
• Updated ServersideInputMultipleOtherStrategy and related data classes to accept and process the new strategy types and arguments, including batch strategies.

Core Logic and Sampling

• Extended core.py to support batch sampling. Now, get_next_points can handle the new strategies, delegating to the appropriate selection logic based on the strategy type.
• Added new utility functions in strategies.py for batch_sampling supporting both liar and believer batch strategies.

Fixes and Minor Improvements

• Fixed initialize_workflow to only train a model if both dataset_x and non-empty dataset_y are provided, preventing unnecessary or invalid model training.
• Improved numerical stability in normalization by enforcing a minimum standard deviation (1e-12) in _transform_Y_params.

These changes collectively improve the flexibility, performance, and correctness of the DIAL service's strategy selection and data handling logic.