GeezerVison

Follow-up to gh-95555 (#151969).

A negated multi-range property such as \P{ASCII} or \P{Pattern_Syntax} was rejected inside a character class. Such a property compiles to a complemented charset, so it could not simply be flattened into the member union. It is now set aside and alternated in with the other members:

• [\P{ASCII}] alone is just the negated charset;
• [\P{ASCII}abc] becomes [abc] | [^ASCII].

Leading ^ (De Morgan) and double negation ([^\P{ASCII}] = ASCII) are handled by the existing set-complement code, and it composes with the set operations: [\w--\P{ASCII}] is the ASCII word characters.

Category-backed \P{...} (e.g. \P{Lu}, \P{digit}), which is a single CATEGORY opcode, always composed fine; only the multi-range properties were affected.

No news entry: \p{...} is unreleased (.. versionadded:: next), so the lifted restriction never shipped.

• Issue: re should support \p{...} character properties #95555