Fix the heap snapshot diff for a memory leak

A Node.js service's memory climbs until it runs out every few days. An AI blamed a listener leak with high confidence. Before you refactor hot-path code on that guess, you diff two heap snapshots taken an hour apart and rank the object types that actually grew.

An assistant wrote rank_heap_growth(labels, before_counts, after_counts, min_growth). It is in the editor. It has a bug. Find it and fix it.

Input and output:

• labels: one object type per row, such as Closure. A label may appear in more than one row.
• before_counts, after_counts: the retained count for each row in the first and second snapshot. Either list may be shorter than labels.
• min_growth: the noise floor.
• Return "<label>=<growth>" strings for the suspects.

Rules the tests check:

• A row's growth is its after count minus its before count. A count missing because its list is too short counts as 0.
• Add up the growth of all rows with the same label.
• A label is a suspect only when its total growth is greater than 0 and at least min_growth. A label that stayed flat or shrank is never a suspect, whatever min_growth is.
• Sort the suspects by growth, largest first. Break ties by label in ascending order.