No cross-codec (Zinc⇄JSON) round-trip tests for DateTime

[jscott3201]

Problem

All existing DateTime round-trip tests are single-codec (encode and decode with the same codec):

• haystack-core/src/codecs/json/v3.rs: datetime_roundtrip, datetime_utc_roundtrip (JSON v3 -> JSON v3)
• haystack-core/src/codecs/json/v4.rs: datetime_roundtrip, datetime_utc_roundtrip (JSON v4 -> JSON v4)
• haystack-core/src/codecs/zinc/parser.rs: roundtrip_datetime, roundtrip_datetime_utc (Zinc -> Zinc)

The integration tests that do exercise cross-codec chains (haystack-core/tests/e2e.rs::codec_round_trip_all_formats and ::multi_codec_grid_fidelity, which encode via Zinc then decode and re-encode via JSON) only cover Ref, Str, Number, Bool, and Marker columns — no test entity in either function sets a Kind::DateTime tag. No test anywhere decodes a DateTime from Zinc and re-encodes/decodes it via JSON (or vice versa) and asserts the offset and tz_name are preserved.

Impact

DateTime is exactly the kind most prone to cross-codec interop drift (Niagara/SkySpark have historically disagreed on offset formatting, Z vs +00:00 for UTC, and tz-name suffix handling), and it's also the kind covered by the DST-resolution gap (see the separate tz issue). Without a cross-codec conformance test, a regression in v3/v4 JSON or Zinc DateTime formatting that breaks Niagara/SkySpark interop could land without any test catching it, since the same-codec round-trip tests exercise the encoder and its own decoder against each other rather than against a second implementation.

Proposed direction

Add a datetime_cross_codec_roundtrip-style test to haystack-core/tests/e2e.rs that builds an HDict with a Kind::DateTime (non-UTC, e.g. New_York with a nonzero offset), round-trips it through each ordered pair of supported codecs (Zinc -> JSON v3, JSON v3 -> Zinc, Zinc -> JSON v4, JSON v4 -> Zinc), and asserts both the dt and tz_name fields survive unchanged.

References

Project Haystack DateTime encoding (Zinc and JSON v3/v4 grid formats).

[jscott3201]

Fixed in #14 (merged to dev as d414d05).

JSON v3's backwards offset scan is replaced with a split on the first space, which fixes the Z UTC form and also removes a reachable panic (t:2024-06-30T12:00:00 -€abcdef sliced mid-codepoint inside find_offset_end, on the decode_grid path).

Cross-codec conformance went further than the issue asked: all four codecs now default an absent tz name to UTC, and both tz-name character sets accept + for the Etc/GMT+N family — Zinc had been truncating GMT+5 to GMT while its encoder wrote GMT+5 back out, silently changing the zone on every round trip.

14 unit tests and a strengthened cross-codec matrix, all mutation-verified.

Closing manually: Closes #N is only honoured for merges into the default branch, and this targeted dev.