Wire integrated dashboard renderer (src/integrated_report.py)

revert: remove post-processing, use raw model output — works best for majority of images

fix: iterative necrotic recovery (3x30px) — bridges gap to necrotic toes

feat: necrotic tissue recovery — dark regions adjacent to foot reclassified as ulcer

fix: shadow filtering + adaptive ITA trimming + increased periulcer dilation (60px) — reduces Fitzpatrick overestimation

fix: preserve dark necrotic tissue connected to foot, only remove isolated dark blobs

feat: postprocess segmentation — largest component + dark pixel exclusion + morphological cleanup

feat: lighting quality assessment for Fitzpatrick estimation

Fix: load 30 selected features from JSON instead of guessing from model

Fix: extract feature names from XGBoost model to match training feature selection (30 vs 63)

Add pipeline code, PWAT models, and Gradio app