Traverse and Transform Nested JSON

lululemon product attributes and order metadata can contain deeply nested JSON objects and arrays. Implement an algorithm that transforms this JSON into a flat dictionary containing one entry for every non-null primitive value.

Formal Specification

Implement flatten_metadata(data), where data is a JSON-compatible object containing dictionaries, lists, strings, numbers, booleans, and null. Return a dictionary mapping each leaf's path to its primitive value.

Use these path rules:

1. Dictionary keys are separated by ..
2. Array elements use zero-based bracket notation, such as items[0].sku.
3. In dictionary keys, escape \\ as \\\\ and . as \\. before joining paths.
4. Omit null values and empty dictionaries or lists.
5. The input root is always a dictionary. Dictionary key order does not affect correctness.
6. The algorithm must use an explicit stack rather than Python recursion so that deeply nested metadata does not overflow the call stack.