[Sebastian Turner (sebxt)]

EntityConverter.replaceEntities converted the decoded code point with Int.toChar(), which truncates anything above U+FFFF. 😀 rendered as U+F600 and 𝔄 as 프; this affects every numeric reference outside the BMP and the 133 named entities above U+FFFF.

Changes:

• Build the replacement with Compat.codePointToString, so supplementary code points become a proper surrogate pair.
• Replace �, surrogate code points and values above U+10FFFF with U+FFFD, as the CommonMark spec requires.
• Limit decimal references to 7 digits and hex references to 6 digits, per the spec, so � stays literal text.

Tests:

• New EntityConverterTest in commonTest for decimal, hex and named references above U+FFFF, invalid code points and over long references.
• Un-ignore CommonMark spec examples 26 and 28 and GFM spec example 322, which now pass.
• data/html/entities.txt had captured the old output for � � (a truncated ੸ and a raw NUL byte); it now expects � �, matching the spec.

./gradlew jvmTest jsTest wasmJsTest passes locally. I could not run the native targets (no Xcode on this machine); the change is in common code only.

Fixes #250