fix: escape IRIREF-forbidden characters and full control range in Writer (#659)

[jeswr]

🤖 Filed by @jeswr's PSS agent (Claude); @jeswr runs multiple agents on this account — the SPARQ agent is a separate party in these discussions.

Draft — addresses #659. N3Writer emitted NamedNode values verbatim between angle brackets, so a value containing an IRIREF-forbidden character (>, <, ", {, }, |, ^, backtick, \, space, or a C0 control) closed its own <…> and injected forged triples (repro 1 in the issue). Separately, literal escaping only covered U+0000–U+0019, so U+001A–U+001F, DEL (U+007F) and the C1 block (U+0080–U+009F) passed through raw (repro 2).

The fix (src/N3Writer.js):

1. IRIs — _encodeIriOrBlank (and @prefix lines) escape the IRIREF-forbidden set (U+0000–U+0020 and < > " { } | ^ ` \) as UCHAR (\uXXXX/\UXXXXXXXX), via a dedicated replacer because IRIREF permits only UCHAR escapes, not ECHAR.
2. Literals — _encodeLiteral widens the escaped ranges to the full C0 plus DEL and C1 (U+0000–U+001F, U+007F–U+009F).

Why escape (UCHAR) rather than percent-encode or throw: percent-encoding changes IRI identity (> → %3E doesn't reverse), whereas UCHAR escaping round-trips byte-identically; a valid IRI contains no forbidden characters, so well-formed data is byte-for-byte unchanged; and re-reading escaped malformed input with N3's own strict parser rejects the invalid IRI instead of materialising an injected quad. If validate-and-throw is preferred instead, that's a small swap — the main reason this is a draft. (DEL/C1 are grammar-permitted raw inside an IRIREF, so they're only escaped in literals; escaping them in IRIs too is a one-line range addition if wanted.)

Tests: a Writer term-serialisation safety (#659) block covers the issue's repros (re-parsing never yields the forged quad, an injected newline can't start a statement), every forbidden character emitted as UCHAR, valid IRIs (%20, query, fragment, punycode host, urn:, did:) byte-for-byte unchanged, astral \U… behaviour preserved, the full C0+DEL+C1 set in literals, and a hostile @prefix. npm test green (6512 tests, 100% coverage on N3Writer.js), lint clean.

Closes #659

[jeswr]

We need to discuss the following RDF/JS assumptions before this PR is merged:

• Do we assume named node IRI's should be pre-escaped

We need to discuss the following N3.js assumption:

• Do we assume prefixes to be sanitised

The following N3.js assumption is known: We expect named nodes to be valid RDF/JS terms.