refactor: solidify startup UX and engine-aware preprocessing

- Slim core deps: move ML stack to optional extras (paddle/tables/figures/scientific/full)
- Lazy settings proxy with config search paths (env var, cwd, user dir)
- New commands: init, demo, setup [basic|full], first-run guard on run/watch
- Engine-aware preprocessing: Paddle gets original image (fixes dark mode 0.83->0.95)
- Results table shows Skipped count; lazy run-dir creation
- kg_ocr marked experimental with extra, Docker defaults with OCR_PIPELINE_CONFIG
- 25/25 tests, ruff clean

tests/test_startup.py

@@ -0,0 +1,116 @@
+from __future__ import annotations
+
+import subprocess
+import sys
+from pathlib import Path
+
+from typer.testing import CliRunner
+
+from ocr_pipeline import config as config_module
+from ocr_pipeline.cli import app
+
+runner = CliRunner()
+
+SRC_DIR = Path(__file__).parent.parent / "src"
+
+
+def _isolate_config(tmp_path: Path, monkeypatch) -> None:
+    """Pretend no config exists anywhere."""
+    monkeypatch.delenv(config_module.CONFIG_ENV_VAR, raising=False)
+    monkeypatch.chdir(tmp_path)
+    monkeypatch.setattr(config_module, "user_config_dir", lambda: tmp_path / "no-such-dir")
+
+
+def test_find_config_file_prefers_env_var(tmp_path: Path, monkeypatch) -> None:
+    env_cfg = tmp_path / "env.yaml"
+    env_cfg.write_text("ocr:\n  engine: tesseract\n", encoding="utf-8")
+    cwd_dir = tmp_path / "cwd"
+    cwd_dir.mkdir()
+    (cwd_dir / "config.yaml").write_text("{}", encoding="utf-8")
+    monkeypatch.setenv(config_module.CONFIG_ENV_VAR, str(env_cfg))
+    monkeypatch.chdir(cwd_dir)
+    assert config_module.find_config_file() == env_cfg
+
+
+def test_find_config_file_uses_cwd(tmp_path: Path, monkeypatch) -> None:
+    monkeypatch.delenv(config_module.CONFIG_ENV_VAR, raising=False)
+    (tmp_path / "config.yaml").write_text("{}", encoding="utf-8")
+    monkeypatch.chdir(tmp_path)
+    assert config_module.find_config_file() == tmp_path / "config.yaml"
+
+
+def test_find_config_file_none_when_unconfigured(tmp_path: Path, monkeypatch) -> None:
+    _isolate_config(tmp_path, monkeypatch)
+    assert config_module.find_config_file() is None
+
+
+def test_import_does_not_read_cwd_config(tmp_path: Path) -> None:
+    (tmp_path / "config.yaml").write_text("{invalid yaml: [", encoding="utf-8")
+    result = subprocess.run(
+        [sys.executable, "-c", "import ocr_pipeline.config; print('ok')"],
+        cwd=tmp_path,
+        capture_output=True,
+        text=True,
+        env={"PYTHONPATH": str(SRC_DIR), "PATH": "/usr/bin:/bin"},
+    )
+    assert result.returncode == 0
+    assert "ok" in result.stdout
+
+
+def test_settings_proxy_delegates() -> None:
+    from ocr_pipeline.config import settings
+
+    assert settings.ocr.engine in {"tesseract", "paddleocr", "auto"}
+    assert repr(settings)
+
+
+def test_init_writes_local_config(tmp_path: Path, monkeypatch) -> None:
+    _isolate_config(tmp_path, monkeypatch)
+    result = runner.invoke(app, ["init", "--local"])
+    assert result.exit_code == 0
+    content = (tmp_path / "config.yaml").read_text(encoding="utf-8")
+    assert "paths:" in content
+    assert "tesseract" in content
+
+
+def test_init_refuses_to_overwrite(tmp_path: Path, monkeypatch) -> None:
+    _isolate_config(tmp_path, monkeypatch)
+    (tmp_path / "config.yaml").write_text("existing: true\n", encoding="utf-8")
+    result = runner.invoke(app, ["init", "--local"])
+    assert result.exit_code == 1
+    assert (tmp_path / "config.yaml").read_text(encoding="utf-8") == "existing: true\n"
+
+
+def test_run_guard_without_config_or_input_dir(tmp_path: Path, monkeypatch) -> None:
+    _isolate_config(tmp_path, monkeypatch)
+    result = runner.invoke(app, ["run"])
+    assert result.exit_code == 2
+    assert "ocr-pipeline init" in result.output
+    assert "ocr-pipeline demo" in result.output
+
+
+def test_watch_guard_without_config(tmp_path: Path, monkeypatch) -> None:
+    _isolate_config(tmp_path, monkeypatch)
+    result = runner.invoke(app, ["watch"])
+    assert result.exit_code == 2
+
+
+def test_setup_dry_run_prints_steps_without_executing() -> None:
+    result = runner.invoke(app, ["setup", "full", "--dry-run"])
+    assert result.exit_code == 0
+    assert "uv sync" in result.output or "pip install" in result.output
+
+
+def test_setup_rejects_unknown_target() -> None:
+    result = runner.invoke(app, ["setup", "everything", "--dry-run"])
+    assert result.exit_code == 2
+
+
+def test_demo_image_bundled_and_valid() -> None:
+    from PIL import Image
+
+    from ocr_pipeline.demo import demo_image_path
+
+    with Image.open(demo_image_path()) as img:
+        assert img.size[0] >= 800
+        assert img.format == "PNG"