Add comprehensive model-free unit coverage across animl-lite core modules

[Copilot]

This PR adds a broad, fast unit test suite for animl-lite focused on core business logic and utility behavior, while avoiding external model/GPU dependencies. It introduces dedicated coverage for utilities, file management/data structures, export formats, ReID distance metrics, and pipeline orchestration via mocks.

• Test harness + dependency isolation

  • Added tests/conftest.py with reusable fixtures and import-time stubs for optional heavy dependencies (onnxruntime, exiftool).
  • Added shared manifest/image fixtures to standardize test setup and keep tests deterministic.

• Utilities + constants (tests/test_utils.py)

  • Covered version reporting, softmax properties, model/type constants, ONNX device/provider selection logic, EXIF/ONNX capability checks, and bbox helper transforms.

• File management + data structures (tests/test_file_management.py)

  • Covered manifest discovery/filtering, recursive/non-recursive behavior, station/camera depth handling, file IO helpers (csv/json/yaml), prompt-driven file checks, WorkingDirectory, active_times, sequence_calculation, ManifestGenerator, and manifest_dataloader.

• Export formats (tests/test_export.py)

  • Added coverage for folder export/update/remove flows and output structure validation for COCO, TimeLapse, and MegaDetector exports (including required-column and edge-case behavior).

• ReID distance metrics (tests/test_reid_distance.py)

  • Added numerical and validation tests for remove_diagonal, euclidean/cosine distance functions, metric dispatch, and batched/non-batched distance parity.

• Pipeline orchestration (tests/test_pipeline.py)

  • Added from_paths/from_config branch coverage with mocked detector/classifier/video/export/visualization dependencies, including detect-only, sequence/single classification, cached detections, and custom detector class-map paths.

• Repo hygiene for test delivery

  • Updated .gitignore to stop ignoring tests/, so the new suite is tracked in source control.

# Example: model-free pipeline orchestration test pattern
monkeypatch.setattr(pipeline.detection, "load_detector", lambda *_a, **_k: object())
monkeypatch.setattr(pipeline.classification, "classify", lambda *_a, **_k: np.array([[0.8]]))
out = pipeline.from_paths("/tmp/images", "det.onnx", "cls.onnx", detect_only=True)
assert len(out) == 1