Phase 7: ground-truth export (JSONL + stats) + CLI tool

- GET /api/v1/ground-truth/export  streaming JSONL (approved_only,
  since, until, has_corrections, limit)
- GET /api/v1/ground-truth/stats   total / approved / corrections
  counts + top-N most-corrected field paths
- python -m ocr_sprint.tools.export_ground_truth  operator CLI with
  the same filters + optional --print-stats
- Ground-truth sample reconstructs the pipeline's original output by
  replaying job_corrections in reverse
- docs/ground-truth-format.md    schema + fine-tuning guidance
- 17 new tests (service replay, endpoint filters, CLI)
- 201 total tests passing, ruff / mypy --strict clean

Co-Authored-By: adrian kuman firmansah <adriancuman@gmail.com>

src/ocr_sprint/main.py

@@ -7,7 +7,7 @@ from fastapi import FastAPI
 from ocr_sprint import __version__
 from ocr_sprint.api.errors import register_error_handlers
 from ocr_sprint.api.metrics import MetricsMiddleware, metrics_endpoint
-from ocr_sprint.api.routes import documents, health
+from ocr_sprint.api.routes import documents, ground_truth, health
 from ocr_sprint.config import get_settings
 from ocr_sprint.db import models as _models  # noqa: F401  (register ORM tables)
 from ocr_sprint.db.base import Base, get_engine
@@ -43,6 +43,7 @@ def create_app() -> FastAPI:
     app.add_middleware(MetricsMiddleware)
     app.include_router(health.router, prefix="/api/v1")
     app.include_router(documents.router, prefix="/api/v1")
+    app.include_router(ground_truth.router, prefix="/api/v1")
     app.add_api_route("/metrics", metrics_endpoint, methods=["GET"], include_in_schema=False)
     return app