Feature/grace wasserstein uncertainty bootstrap

[graxia0322]

Edited 01_data_import.r, 02_wasserstein_core.r, 03_visualization.r and wasserstein_bootstrap_analysis.r to encompass up to 3x3x3 formulation x device resistance x pressure drop inhaler data in each script in a flexible manner. Key features of each script:
01_data_import.R
Data Structure:
RODOS: Traditional folder-based structure (formulation folders with replicate CSVs)
INHALER: Flat folder with metadata embedded in CSV headers

Automatic Metadata Extraction:
Device Resistance: Auto-detects "low", "medium", "high" from Device column (case-insensitive)
Pressure Drop: Extracts numeric values from pressure_drop/pressure-drop columns
Formulation ID: From folder names (RODOS) or formulation_id column (INHALER)
Replicates: From filename pattern (RODOS) or auto-assigned by timestamp (INHALER)

Flexible Factor Handling:
Works with any combination: 3×1×1, 3×2×3, 3×3×3, etc.
Handles missing factors (e.g., single device level)
Validates all extracted metadata with informative warnings

Smart Replicate Assignment:
INHALER: Groups by formulation + device + pressure, then assigns rep1, rep2, rep3 by timestamp
RODOS: Extracts from filename using regex pattern

Output:
data_v2/tidy/standardized_data_with_conditions.csv

02_wasserstein_core.R
Nested Looping for All Conditions:
Outer loop: Formulations
Inner loop: Device × Pressure combinations
Calculates W1 for each unique condition vs formulation-specific RODOS reference

Correct Pooling Methodology (pool + average across particle size grid):
RODOS: Pools ALL replicates once per formulation (reference is condition-independent)
INHALER: Pools replicates separately for each resistance×pressure drop combo
Follows correct statistical approach: pool first, then calculate W1

Flexible Condition Filtering:
pool_replicate_cdfs() accepts optional device_resistance and pressure_drop parameters
When NULL, pools across all conditions (for RODOS)
When specified, filters to specific condition (for INHALER)

Comprehensive Validation:
Reports W1 results with device/pressure identifiers
Validates data availability for each condition

Output:
results_v2/wasserstein_results.csv

03_visualization.R
Dynamic Facet Grid Generation:
Auto-creates device_resistance (rows) × pressure_drop (columns) facet grids
Repeats RODOS reference curve in every facet for comparison
Scales plot dimensions automatically based on number of facets

Intelligent Factor Ordering:
Device: Natural order (low → medium → high) if those levels exist
Pressure: Numeric sorting (1 kPa → 2 kPa → 4 kPa)
Falls back to alphabetical if non-standard levels

Four Plot Types:
Individual Formulation PDFs: One PDF per formulation showing all device×pressure grids
Overlay PDF: All INHALER conditions colored by formulation
CDF by Device: Faceted by device resistance levels
CDF by Pressure: Faceted by pressure drop levels
W1 by Device/Pressure: Bar charts faceted by experimental factors

Output:
figures_v2/*_comparison_faceted.pdf (one per formulation)
figures_v2/all_inhaler_overlay.pdf
figures_v2/CDF_by_device.pdf
figures_v2/CDF_by_pressure.pdf
figures_v2/W1_by_device.pdf
figures_v2/W1_by_pressure.pdf

wasserstein_bootstrap_analysis.R
Factor-Aware Bootstrap Resampling:
get_replicate_cdfs() accepts device_resistance and pressure_drop filters
Bootstraps each device×pressure combination separately
Generates uncertainty estimates for all conditions, not just formulations

Three-Way Effect-to-Noise Analysis:
Formulation Effect: Variability across formulations (held device/pressure constant)
Device Effect: Variability across device levels (held formulation/pressure constant)
Pressure Effect: Variability across pressure levels (held formulation/device constant)
Each ratio compares effect magnitude to bootstrap measurement noise

Defensive Factor Checking:
Detects if factors exist in data before calculating effect-to-noise ratios
Skips device/pressure effects if only 1 level present (e.g., 3×1×1 design)
Works from simple (3×1×1) to complex (3×3×3) designs

Flexible Faceted Visualizations:
Bootstrap Analysis PDF: Shows CI, SE, and precision for all conditions with device×pressure facets
Effect-Noise Analysis PDF:

• Panel A: W1 values by formulation (collapsed across conditions)
• Panel B: Effect vs Noise comparison for all three factors
• Panel C: Measurement precision by formulation
• Panel D: Effect-to-noise ratio bars with color-coded signal strength
  Auto-adjusts plot dimensions based on number of facets

Smart Faceting Logic:
Only facets when factors have >1 level
Uses facet_grid() for 2D layouts when both device and pressure vary
Uses facet_wrap() for 1D layouts when only one factor varies
No faceting for simple formulation-only designs

Output:
results_v2/bootstrap_results.csv (W1 mean, SD, 95% CI for each condition)
results_v2/effect_noise_ratios.csv (3 rows: formulation, device, pressure effects)
figures_v2/bootstrap_analysis.pdf (faceted bootstrap diagnostics)
figures_v2/effect_noise_analysis.pdf (multi-panel effect-to-noise summary)