Parameter optimization (WIP)

[Lars800]

Partial implementation of Optimization funcutionality for Openstef Beam.

This module introduces the BenchmarkOptimizer and DatasetOptimizer.
Their shared components are implemented in the OptunaOptimizer class.

• BenchmarkOptimizer optimizes a TargetProvider
• DatasetOptimizer optimizes a VersionedTimeseriesDataset (ground truth, predictors)

To initialize an optimizer we need to specify an Optimizer config consisting of the following components:

• ParameterSpace
  - Hyperparams-like object that generalizes parameters to parameter-ranges (Floats, Ints, Catagorical)
  - Implements a make_optuna() method that produces Optuna compatible dictionaries
  - Default Presets for LGBM(Linear), XGBoost and GBLinear
• BacktestForecasterConfig - Config for the OpenstefBacktestForecaster
• BacktestConfig - Config for BacktestPipeline
• ForecastingWorkflowConfig - Base config where hyperparameters will be overwritten
• OptimizationMetric - Bundles a MetricProvider with a direction to create the optimization goal.
• Cache dir - place to store trial results for efficient multiprocessing

Potential improvements:

• Easier configuration
• Parameterising pre-and post processing
• Make compatible with openstef-meta (Especially forecast combiners)