Implement a function micro_f1_score(y_true, y_pred) that computes the micro-averaged F1 score for multi-label classification. Inputs are two 2D NumPy arrays of shape (n_samples, n_labels) containing binary values (0 or 1), where each row is a binary indicator vector representing the labels assigned to a sample.

The micro-averaged F1 aggregates true positives, false positives, and false negatives across all labels (and samples) into a single global count, then computes precision, recall, and F1 from those totals.

Return a single float (the micro F1 score). Do not round inside your solution. If the denominator of precision, recall, or F1 is zero, return 0.0 for that quantity.