It's possible that 'retrieve a preferred prefix string' returns a wrong prefix for the specified namespace

https://w3c.github.io/DOM-Parsing/#dfn-retrieving-a-preferred-prefix-string
WPT domparsing/XMLSerializer-serializeToString.html has a test case for this behavior."Check if an attribute with namespace and no prefix is serialized with the nearest-declared prefix even if the prefix is assigned to another namespace."

const root = (new DOMParser()).parseFromString('<el1 xmlns:p="u1" xmlns:q="u1"><el2 xmlns:q="u2"/></el1>', 'text/xml').documentElement;
root.firstChild.setAttributeNS('u1', 'name', 'v');
console.log((new XMLSerializer()).serializeToString(root));

If we follow the current specification, 'retrieve a preferred prefix string' for name attribute choses q prefix because q is in the list for u1 namespace. However, namespace for q is u2 on the element. The serialization result would be <el1 ...><el2 xmlns:q="u2" q:name="v"/></el1>

Before we add prefix and namespace to a map, we need to remove the prefix associated to another namespace from the map.

Start with the DOM-Parsing specification section on “retrieving a preferred prefix string” and the WPT case in domparsing/XMLSerializer-serializeToString.html. Reproduce the XMLSerializer example, then trace how prefixes are selected and mapped during serialization. Done means the specified namespace case serializes with a prefix bound to the correct namespace and the WPT test passes.