fix: prevent unparse crash on split surrogate entities

[sabbirexist]

• Fixes a UnicodeDecodeError in HTML and Markdown unparse when an entity boundary falls inside a UTF-16 surrogate pair.
• Adds clamp_to_code_point() to safely widen misaligned boundaries to the complete Unicode code point.
• Keeps all valid/aligned entity behavior unchanged.
• Adds 13 regression tests covering the affected cases.
• Full test suite: 14862 passed, 35 skipped.

[rjriajul]

Thanks for this. The half-emoji crash is real, and the new tests catch it (7 of them fail on dev and pass on this branch).

One case I hit while testing: when two entities overlap after the widening, the Markdown output puts the tags in the wrong order. Examples:

• a😀b with bold at offset 1 length 1 and italic at offset 2 length 1 gives a**__😀**__b
• 😀x with italic at offset 0 length 1 and bold at offset 1 length 1 gives __**😀__**x

HTML comes out fine for both. Could you add a test for overlapping entities in Markdown and make sure the output stays well formed? Let me know when it's pushed and I'll take another look.