Implement fn:parse-ietf-date with a hand-written parser

[ajbufort]

Fixes #36.

The previous implementation delegated to chrono's parse_from_rfc2822, which rejects most of the formats the spec requires (the asctime layout, two-digit years, named timezones, single-digit fields, liberal whitespace) and accepts some it forbids (three-digit years). 51 conformance tests were filtered as known failures.

Following the direction in #36, this replaces it with a hand-written recursive descent parser for the exact grammar in F&O 3.1 section 9.9.1, adding no dependencies to xee-interpreter.

What's covered

• Both date layouts (datespec S time and asctime), optional/ignored dayname, comma and hyphen separator forms
• Two-digit years (+1900), four-digit years taken as given (e.g. 0070)
• Named timezones (UT/UTC/GMT/EST/...), offsets with and without colons (-05, -05:, -5:00, -500, -0500), the parenthesized-and-ignored tzname after an offset, and the +/-14:00 range check
• Fractional seconds, 24:00[:00] rolling over to the next day, calendar validation (Feb 29, day 00/32), with FORG0010 on any violation
• Only x09|x0A|x0D|x20 count as whitespace, per the grammar's S production

Testing

• New unit tests in library/ietf_date.rs cover the spec examples, the whole conformance truth table, and edge cases the suite doesn't reach (offset boundary +14:00/+14:01, half-hour offsets in every lexical form, tab/CR/LF whitespace, 24:00:00.5 and a nonzero fraction beyond nanosecond precision, leap seconds, non-ASCII whitespace at the edges and interior, lexical year extremes 0001/9999 including 24:00 rollover)
• A small deterministic mutation-fuzz test (50k inputs derived from valid examples) guards against panics on arbitrary input
• All 105 tests in fn/parse-ietf-date.xml and the 5 fo-test-fn-parse-ietf-date tests in misc now pass: the XPath check run goes from 20221 to 20272 passed, filters updated via the update/check flow from hacking.md, no regressions in the XPath or XSLT suites (verified in release and in debug mode, as CI runs it)
• cargo fmt, cargo clippy --all-targets --all-features -- -D warnings, and cargo test are all clean

Notes

• The old stub trim()ed all Unicode whitespace before parsing; this parser accepts exactly the grammar's S set (x09|x0A|x0D|x20) at the edges, so e.g. a trailing U+00A0 is now FORG0010 where it previously parsed. I read the spec's whitespace note as deliberate about those four characters, so the stricter behavior is pinned by tests - easy to relax if you'd rather keep the lenient trim.
• parse-ietf-date("Dec 31 24:00 9999") correctly yields the year-10000 xs:dateTime, which exposes a pre-existing serializer quirk: five-digit years print with a leading + (+10000-...), which the engine's own xs:dateTime cast then rejects. That's reachable on main via xs:dateTime("10000-01-01T00:00:00Z") too - happy to file it separately.

The parser lives in its own module (xee-interpreter/src/library/ietf_date.rs) with parse_ietf_date in datetime.rs delegating to it - happy to restructure if you'd prefer it inline or elsewhere.

Generated with Claude Code