Upload docclass_merged.jsonl with huggingface_hub

KANBAN-084: schema v5 — +insurance_claim class (CMS DE-SynPUF) + clause-level GT (CUAD annotations / MAUD gold)

Card: dedicated source-attribution section (CUAD/MAUD/S-1/Enron original downloads + verified academic references) and full citation block

v4: fuse stratified 110-row correspondence sample from enron-correspondence-dedup; two-config blind/GT parquet layout; legacy JSONL preserved

Docclass merged corpus schema v3 — deterministic train/test split column (KANBAN-074, 700 rows, fp cd652e771035)

KANBAN-076: document conversion provenance (normalized metadata, manifest.txt)

Docclass merged corpus schema v3 — deterministic train/test split column (KANBAN-074, 700 rows, fp cd652e771035)

KANBAN-076 round-2 repair: ship manifest as manifest.txt (no json substring -> loader-invisible)

KANBAN-076 round-2 repair: remove manifest.json.txt (loader ingests ANY *.json* path - canary-proven)

Repair: ship manifest as manifest.json.txt (loader-invisible, human-readable — KANBAN-076)

Repair: remove bare manifest.json (Hub JSON loader ingests it as data rows — KANBAN-076)

Docclass merged corpus schema v3 — deterministic train/test split column (KANBAN-074, 700 rows, fp cd652e771035)

Docclass merged corpus schema v2 — subclasses + filenames on every row (KANBAN-073, 700 rows, fp cd652e771035)

Docclass merged corpus (KANBAN-071, 700 rows, fp 5b682f62ee94)

initial commit