feat: OCR scoring, adaptive fallback, NIK region fallback, relaxed rejection

fix: Use fix_nik_string (no digit-to-digit conversion), add Nama fallback spatial, fix berlakuHingga mapping

refactor: Replace spatial sequence with label-pattern regex extraction (comprehensive)

fix: Add Keyword Guard to spatial sequence extraction to prevent Nama and Alamat mix-ups

Fix duplicate code and implement spatial Y-coordinate layout parsing

Major OCR pipeline upgrade: dual-pipeline, deskew, fuzzy matching, confidence scoring, quality gate

Enhance sequence logic using RT/RW as anchor for full address block

Revert cv2.findContours crop, implement robust sequence fallback for NIK, Nama, Alamat

Add auto-crop using cv2.findContours and update gallery upload note

Cleanup stray punctuation like dots and commas in OCR output

Fix address regex: SINDUJOYONO.8 -> SINDUJOYO NO. 8

Fix address regex, scanNext returns to same input, bulk queue engine

Fix critical OCR bugs: remove global char replacement, revert preprocessing, fix berlakuHingga

Add health check root endpoint

Fix typo berlakuHingga

Update OCR logic

Refine regex for JK, Goldarah, Kewarganegaraan and Agama robust parsing

Implement Hybrid Regex Engine (Value-Based + Sequence-Based Heuristics)

Fix invalid regex character range causing re.error

Enhance OCR accuracy with image upscaling and extract exhaustive KTP fields

Pin paddleocr to stable version 2.9.1 and restore cls parameter

Downgrade paddlepaddle to 2.6.2 to fix OneDNN backend crash on CPU

Remove deprecated cls parameter from ocr.ocr call

Fix HF Docker user permissions and add traceback to error response

Convert grayscale image back to BGR for PaddleOCR compatibility

Fix libgl1 package name for newer Debian versions

Restore HF Space configuration files

Fix OpenCV libGL error and update OCR logic

Upload 3 files

initial commit