fix: preserve negative zero sign when stringifying values

[rajanpanth]

Summary

Every raw-value-to-string conversion in the builder silently drops the sign of -0, corrupting it to 0:

const builder = new XMLBuilder();
builder.build({ a: -0 }); // '<a>0</a>' -- should be '<a>-0</a>'

This affects text values, attribute values, oneListGroup array items, and stopNode raw content — in both preserveOrder modes. XML has no separate int/float syntax, so the sign carries the only signal that a value was a negative float rather than 0, and it's silently lost on a build round trip (e.g. parser.parse(builder.build(x)) no longer matches x for any -0 value).

Root cause: String(val), val.toString(), and '' + val all render JS -0 as "0" — none of them special-case it — and the builder does this raw-to-string conversion at many separate call sites across fxb.js, orderedJs2Xml.js, and util.js.

Changes

• src/util.js: add a small shared valToStr() helper that special-cases Object.is(val, -0), and use it inside safeComment/safeCdata/escapeAttribute instead of their bare String(val).
• src/fxb.js / src/orderedJs2Xml.js: replace every ad hoc '' + val / val.toString() / implicit-ToString raw-value stringification with valToStr(), covering: attribute values, text values, oneListGroup array items, and stopNode raw content, in both preserveOrder: false and preserveOrder: true.
• Tests added alongside each affected code path in spec/j2x_spec.js, spec/j2x_ordered_spec.js, and spec/stopNodes_spec.js (7 new cases total).

Testing

• Full suite: 136/136 passing (129 pre-existing + 7 new), 1 pre-existing pending spec unrelated to this change.
• Manually verified every affected path before/after (text, attribute, oneListGroup, stopNode, CDATA, comment — both preserveOrder modes) using a throwaway script, then encoded the cases as committed tests.
• npm run lint reports the same pre-existing errors with and without this change (unrelated files/lines; confirmed via git stash) — this PR introduces no new lint issues.