Add pytest coverage for video_processing, detection, and classification modules

[Copilot]

Adds missing test coverage for the three core inference modules (video_processing, detection, classification), following the existing pytest structure and conftest stub pattern.

New test files

• tests/test_video_processing.py — get_images, get_videos, extract_frames (error cases, image-only, mixed, file output), _count_frames (missing file, frame list shape, fps mode)
• tests/test_detection.py — load_detector (missing file, model type), parse_detections (columns, empty detections, threshold filtering, manifest merge, error cases, file output), get_animals/get_empty (label + category int fallback), _convert_detections (no-detection path)
• tests/test_classification.py — load_classifier (with/without classes arg), load_class_list, classify (DataFrame/string/list inputs, missing column, file output, invalid input), single_classification (prediction column, empties, best=True, empty animals)

conftest.py

Extended _install_optional_stubs() to stub cv2 when OpenCV is not installed, consistent with existing onnxruntime and exiftool stubs. The _FakeCapture returns 30 frames at 10 FPS so video extraction tests run without real video files.

# Example: video extraction works with stub cv2 and a zero-byte placeholder
def test_extract_frames_includes_videos(video_manifest):
    result = extract_frames(video_manifest, frames=3, parallel=False)
    assert "frame" in result.columns and len(result) >= 1

• Fixes add tests for video, detection, classification #18