Fix semver:eq#3 ignoring $coerce parameter

[joewiz]

Summary

• semver:eq#3 was calling semver:compare#2 instead of semver:compare#3, causing the $coerce argument to be silently ignored. Calling semver:eq("1.0", "1.0.0", true()) would throw an error instead of returning true().
• semver:eq#2 is updated to delegate to semver:eq($v1, $v2, false()), consistent with all other arity-2 comparison functions (lt, le, gt, ge, ne).
• Two regression tests added to compare.xqm covering the equal and not-equal cases with coercible non-SemVer strings.

Test plan

• stc:eq-coerce-non-semver — semver:eq("1.0", "1.0.0", true()) returns true()
• stc:eq-coerce-non-semver-not-equal — semver:eq("1.0", "2.0.0", true()) returns false()

🤖 Generated with Claude Code

[joewiz]

For the record, here's the log of the session that produced this PR:

https://gist.github.com/joewiz/0d90cae455ca015bac2a8d2a05eb202c

[line-o]

Looking good.
Would it make sense to have the opposite test as well?

declare
    %test:assertFalse
function stc:eq-non-semver-without-coerce-fails() {
    semver:eq("1.0", "1.0.0", false())
};

[joewiz]

@line-o Thanks! The attempt to parse 1.0 without coercion would cause an error to be thrown. We already test the library's ability to detect "incomplete" SemVer strings (and thus throw that error) here: https://github.com/eXist-db/semver.xq/blob/master/src/test/xquery/validate-and-parse.xqm#L106. So I don't think we need an extra test here.