Should dataframes with empty predictions have a class recall of None or 0?

This may get superseded by the desire to transition to torchmetrics for evaluate.py, but I am having internal debate about how to format the evaluation dataframe for situations where the model returns no predictions.

related to https://github.com/weecology/DeepForest/issues/412

and uncovered during the PR here: https://github.com/weecology/DeepForest/pull/410

def test_evaluate_empty():
    m = main.deepforest()
    m.config["score_thresh"] = 0.8
    csv_file = get_data("OSBS_029.csv")
    root_dir = os.path.dirname(csv_file)
    results = m.evaluate(csv_file, root_dir, iou_threshold = 0.4)
    
    #Does this make reasonable predictions, we know the model works.
    assert np.isnan(results["box_precision"])
    assert results["box_recall"] == 0
    
    df = pd.read_csv(csv_file)
    assert results["results"].shape[0] == df.shape[0]

Start with evaluate.py and the test_evaluate_empty case using OSBS_029.csv. Review related issue 412 and pull request 410 to determine the expected representation for empty predictions and whether the existing precision, recall, and dataframe-shape assertions reflect it. Done means the behavior is decided and the relevant evaluation test passes.