test(yolo-tracker): Define CPU vs GPU test strategy and markers

[rogermt]

Problem

After DRY refactor and team_classification removal, we need clear CPU/GPU test separation:

1. CPU-only tests should NOT require models

   • Fast tests (<1s)
   • No YOLO model loading
   • Mock dependencies
   • Run by default in CI

2. GPU tests require actual models

   • Slow tests (10-30s)
   • Require RUN_MODEL_TESTS=1
   • Require downloaded model files
   • Run only on GPU environments (Kaggle/Colab)

Current Issues

• Test markers inconsistent across files
• Some tests incorrectly marked as GPU-only
• MODEL_PATH exports break when models don't exist (CPU environment)
• test_models_directory_structure.py fails on CPU (expects actual models)

Solution

Test Configuration Pattern

All GPU tests must follow:

import os
import pytest
from tests.constants import RUN_MODEL_TESTS

@pytest.mark.skipif(
    not RUN_MODEL_TESTS,
    reason="Set RUN_MODEL_TESTS=1 to run (requires GPU/models)"
)
def test_requires_gpu():
    pass

Files to Fix

• test_models_directory_structure.py - MODEL_PATH imports fail on CPU
• test_video_model_paths.py - Same issue
• test_plugin.py - analyze() needs GPU models
• test_plugin_all_detections.py - Same
• Inference modules - Don't export MODEL_PATH (causes import errors on CPU)

Definition

CPU Tests: Run always

• Unit tests with mocks
• Config tests
• Utils tests

GPU Tests: Run with RUN_MODEL_TESTS=1

• Inference tests needing actual YOLO models
• Plugin.analyze() tests
• Model file size validation

Related Issues

• chore(mypy): Add missing type stubs for yaml, sklearn, tqdm, umap #48 (mypy errors - needs reopen)