fix: do not line-end-normalize a CR character reference in content

[spokodev]

Problem

A character reference resolving to a carriage return (
 or 
) in element text content is wrongly converted to a line feed.

parseXml('<a>&#xD;</a>').root.children[0].text // '\n'  (wrong)
// expected: '\r'

XML 1.0 section 2.11 end-of-line normalization applies only to literal line breaks in the input, not to characters produced by a character or entity reference (see sections 4.6 / 4.1). The reference must be passed through to the application as the literal character it denotes.

The cause: consumeContentReference routes the resolved reference text through addText, which unconditionally calls normalizeLineBreaks. So a resolved 
 is collapsed to \n just like a literal CR in the source would be.

Evidence this is the intended behavior

• The vendored W3C conformance test xmltest/valid/sa/067.xml (<doc>&#13;</doc>) has canonical output <doc>&#13;</doc>, i.e. the CR must survive.
• The existing attribute-side test ("doesn't normalize a character reference for a whitespace character", issue 
 is interpreted as space instead of line feed #6) already asserts this for attributes. The content path was simply never tested, and the attribute path uses a different code route, so the bug stayed hidden.

Fix

Add a normalize parameter to addText (default true) gating the normalizeLineBreaks call, and pass false from consumeContentReference. The CDATA and CharData callers keep the default true, since literal content is subject to section 2.11.

Verification

• New test fails on main ('\n' !== '\r') and passes with the fix.
• Full suite green (1378 passing, up from 1377), 100% coverage maintained.
• W3C conformance suite still green.

References

• https://www.w3.org/TR/2008/REC-xml-20081126/#sec-line-ends
• https://www.w3.org/TR/2008/REC-xml-20081126/#entproc

[rgrove]

Thanks!