XCharacterClassification::parsePredefinedToken() with
KParseType::IDENTNAME and KParseTokens::ANY_LETTER does exactly
that, it parses one identifier name consisting of Unicode letters
and stops at the first non-letter character (additionally '-'
hyphen-minus was allowed as character).

Instead, scan for the first (any Unicode) digit and split there.

Also, copying the whole string to the prefix string if no split
happens was unnecessary, the caller does not use the prefix if no
number was found. Nail this as post condition.

Change-Id: I941f9739b39c36c83b63145e3b762ec558738c1c