Fix indentation when the indent character is not ASCII

[youdie006]

The bug

Indentation (src/writer.rs:670-737) keeps current_indent_len as a byte index into its indents
cache -- current() slices with it -- but grow and shrink moved it by indent_size, which is a count of
characters. The two agree only while the indent character is one byte long.

Writer::new_with_indent(inner, indent_char: u8, indent_size: usize) is public and takes a u8, and
char::from(u8) maps every byte above 0x7F to a character that is two bytes in UTF-8. So the mismatch is
reachable from safe code with a plain byte argument:

// 0xA0 -> U+00A0 NO-BREAK SPACE, two bytes in UTF-8
let mut w = Writer::new_with_indent(Vec::new(), 0xA0, 1);
w.write_event(Event::Start(BytesStart::new("a")))?;
w.write_event(Event::Start(BytesStart::new("b")))?;

thread 'main' panicked at src/writer.rs:710:22:
end byte index 1 is not a char boundary; it is inside '\u{a0}' (bytes 0..2 of string)

With an even indent_size it does not panic, it just indents half as far as asked:

indent_size = 2, expected "<a>\n\u{a0}\u{a0}<b>t</b>\n</a>"
                 actual   "<a>\n\u{a0}<b>t</b>\n</a>"

Separately, Serializer::indent (src/se/mod.rs:819) accepts a char and then throws most of it away:

self.ser.indent = Indent::Owned(Indentation::new(indent_char as u8, indent_size));

'\u{2003}' as u8 is 0x03. Serializing with indent('\u{2003}', 2) produces

"<Outer>\n\u{3}\u{3}<inner>\n\u{3}\u{3}\u{3}\u{3}<t>x</t>\n\u{3}\u{3}</inner>\n</Outer>"

U+0003 is not a legal XML character (XML 1.0 Char excludes C0 controls other than tab, LF and CR), so the
serializer returns Ok with a document that a conformant parser must reject. Any indent character above
U+007F -- an em space, an ideographic space -- goes the same way, and char is the only type the caller can
pass.

The fix

• Add Indentation::with_char(char, usize); Indentation::new(u8, _) becomes a thin wrapper over it, so
  Writer::new_with_indent's public signature is unchanged.
• Add level_len() = indent_size * indent_char.len_utf8(), and use it in grow and shrink so
  current_indent_len stays a byte length.
• In additional(), convert the added level the same way. The value passed in is
  AttributeIndent::WriteConfigured(i.indent_size) (src/writer.rs:533), and the variant's own doc comment
  at src/writer.rs:444 says "Write specified count of indent characters" -- so the added amount is a
  character count and needs the same conversion, applied to the added level only, not to the running total.
• Point Serializer::indent at with_char so the char it already accepts survives.

ASCII is untouched: len_utf8() is 1, and every arithmetic site reduces to what it was.

I kept new_with_indent's u8 rather than widening it to char, since that would be a breaking change for
every caller. Happy to switch it if you would rather fix the signature at the same time.

Verification

New module multi_byte_indent_char in tests/writer-indentation.rs: nested elements, attribute indent at
depth 0 and at depth 1, an odd indent_size (the case that panicked), an ASCII case pinning that nothing
moved, and the serializer case.

• cargo test --all-features passes; so do --no-default-features, --features serialize,
  --features serialize,encoding and --features serialize,escape-html, and
  cargo test --all-features --benches --tests.
• cargo fmt -- --check clean, cargo +1.86.0 check (the MSRV the workflow pins) clean.
• Reverting the whole patch fails 5 of the new tests and no existing ones.
• Two one-line mutations fail disjoint sets, so neither half of the change is dead weight and the
  boundary is pinned from both sides:
  • restoring only Indentation::new(indent_char as u8, ..) in Serializer::indent fails only
    serializer_indent_char_is_not_truncated;
  • over-correcting additional() to (current_indent_len + additional_indent) * len_utf8() fails only
    in_attributes_nested.

That second one is why in_attributes_nested exists. The depth-0 attribute test cannot see the difference --
current_indent_len is 0 there, so both forms give the same answer. It only shows up one level in.

AI disclosure

Written with AI assistance (Claude Code, claude-opus-5). Per the AI policy: I have reviewed the change, I can
explain it without going back to the agent, and I will answer review questions in my own words rather than
pasting model output.

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 55.10%. Comparing base (e00ae5c) to head (491ea01).
⚠️ Report is 76 commits behind head on master.
❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1017      +/-   ##
==========================================
- Coverage   57.31%   55.10%   -2.22%     
==========================================
  Files          46       51       +5     
  Lines       18197    18822     +625     
==========================================
- Hits        10429    10371      -58     
- Misses       7768     8451     +683     

Flag	Coverage Δ
unittests	55.10% <100.00%> (-2.22%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.