Introduce scheduler infrastructure for resource allocation and bracket management.
Support Fixed, BOHB, MFES, and Flatten scheduling strategies.

Add scheduler support with factory pattern and fidelity-aware evaluation.
Enable BOHB/MFES successive halving strategies in generic optimizer.

…trategy interface

- Implement sampling strategies:
  * StandardSamplingStrategy for single-space sampling
  * MixedRangeSamplingStrategy for adaptive compressed/original space sampling

…work (base/expert/shap)

Co-authored-by: Lingching <Elubrazione@users.noreply.github.com>

- Refactor range compression into modular step classes:
  * BaseRangeCompressionStep: Base class for range compression steps
  * BoundaryRangeStep: Simple boundary-based range compression with mixed sampling support
  * SHAPBoundaryRangeStep: SHAP-weighted boundary range compression
  * ExpertRangeStep: Expert-specified range compression

Co-authored-by: Lingching <Elubrazione@users.noreply.github.com>

- Add source_similarities support to dimension and range compression steps:
  * SHAPDimensionStep: Weight SHAP importances by task similarity
  * SHAPBoundaryRangeStep: Use similarity-weighted SHAP values, focus on beneficial samples (SHAP < 0)
  * TunefulDimensionStep: Pass source_similarities to parent class
- Refactor data extraction utilities
- Improve SHAP range compression:
  * Only use beneficial samples (SHAP < 0) for range computation
  * Combine SHAP weights with similarity weights for better parameter selection

- Add core compression framework:
  * CompressionStep base class for extensible compression operations
  * CompressionPipeline for orchestrating multi-step compression workflows
  * OptimizerProgress for tracking optimization state and adaptive updates
- Support adaptive compression updates based on optimizer progress
- Enable point projection/unprojection through compression pipeline

- KDEBoundaryRangeStep: KDE-based range compression with similarity weighting
- QuantizationRangeStep: Quantization-based compression with adaptive updates

- Add SpearmanDimensionStep for correlation-based dimension selection
- Add KPCAProjectionStep for kernel PCA-based projection
- Add bidirectional caching to REMBO/HesBO projection steps

…ons between spaces

- filling strategy initialization
- unproject method for quantization

- Refactor Compressor/__init__.py to use new pipeline architecture
- Add new Compressor base class in Compressor/core/compressor.py
- Enhance CompressionPipeline info
  * Add detailed step-by-step compression statistics
  * Display dimension ratio and effective compression ratio separately
  * Show comprehensive pipeline summary with original/sample/surrogate dimensions
- Improve projection step visualization support

Add visualization functions for compression pipeline analysis including:
- Compression details and range compression visualization
- Dimension reduction tracking across steps
- Parameter importance visualization and heatmaps
- Pipeline flow diagrams
- Compression metadata export

- Parameter filtering -> DimensionSelectionStep
- Value clipping -> RangeCompressionStep
- Parameter filling -> all steps
- Add clipping.py utilities

Each step now fully owns its transformation logic.
Fixes out-of-bounds crash when ranges compress.

…mation

- All compressor types now use convert_config_to_surrogate_space()
- REMBO/HesBO support pseudoinverse approximation for cache misses

Implement progressive compression mode for PeriodicDimensionStep to enable
gradual dimension reduction instead of always
restarting from the original space

- Add UpdateStrategy framework with periodic/stagnation/improvement strategies
- Add ImportanceCalculator with SHAP and correlation-based implementations

- Refactor SHAPDimensionStep to use SHAPImportanceCalculator
- Replace SpearmanDimensionStep with generic CorrelationDimensionStep
- Remove redundant validation checks from calculators
- Update exports to reflect new architecture

- Change source_similarities from List[Tuple] to Dict[int, float] across all steps
- Add source_similarities param to all compress() methods for complete data flow

- Add AdaptiveDimensionStep
  * Integrates with UpdateStrategy framework for dynamic dimension adjustment
  * Handles dimension increase/decrease based on optimization progress

- Enhance UpdateStrategy framework
  * Add compute_new_topk() method to all strategies
  * Implement dimension adjustment logic for each strategy type

- Improve CompressionPipeline
  * Detect dimension increase scenarios and use original_space when needed
  * Simplify project_point/unproject_point to handle dict directly
  * Better handling of progressive vs re-compression scenarios

- Add compression info saving to JSON files with history tracking
- Implement get_step_info() across all compression steps
- Add source_similarities parameter support throughout pipeline

- Track unprojected_space in CompressionPipeline
- Add get_unprojected_space() to Compressor
- Fix BO advisor to use correct target space for unprojection

- unproject to sample space when no transformative step
- return  instead of
- add new interface  to unproject batch configs

- parameter exclusion (allowing experts to exclude certain parameters)
- add expert_params parameter to all dimension step constructors (Adaptive, SHAP, Correlation, Expert)
- refactor _select_parameters() to return all sorted parameters instead of just topk
  - base class now handles topk selection and expert_params merging
  - this allows expert params to be prioritized while method selection supplements
- simplify ExpertDimensionStep to delegate expert param handling to base class

…gement

…itial design

…ce_ratio-aware observation updates

…h flow

… function