PROV-N and PROV-JSON qualified-name lexical handling

[amateescu]

Changes

• The PROV-N parser now accepts the rest of the grammar's PN_CHARS_OTHERS set (@ & + * $ ! % plus percent-encoding) in qualified names, so names like ex:a@b that ProvToolbox emits no longer fail with Expected ')'.
• A new internal QualifiedNameEscaper backslash-escapes PN_CHARS_ESC punctuation in local names on output and decodes it on input. Delimiter punctuation (( ) [ ] ' = , ; :) is escaped everywhere; . only when leading or trailing and - only when leading, per PN_LOCAL, so common dotted names like node.article.body keep their familiar lexical form. The PROV-N serializer escapes these instead of rejecting representable names, and the deserializer decodes them, so a name's canonical local part is the decoded form (matching the RDF/Turtle ground truth) rather than leaking a literal backslash.
• The same escape/decode is applied to PROV-JSON: ProvToolbox carries the PROV-N escaping inside its prefix:local strings, so the JSON serializer escapes and the deserializer decodes, keeping all formats in agreement on the canonical local part.
• The PROV-N string-literal reader gains the missing \b, \f and \' ECHAR sequences, which were previously left verbatim.
• Every qualified-name emission (identifiers, relation endpoints, attribute values, datatypes) in PROV-N, PROV-JSON and PROV-XML now routes through PrefixMinter, so a value whose namespace was never declared still gets a declaration (a root xmlns in XML) instead of unparseable output or an unbound prefix. The minter preserves each name's own namespace boundary, so a versioned local part containing / (e.g. node:42/rev/7) is never re-split.