Parse errors carry line/column + context snippet

[conorbronsdon]

Adds position info to parse errors, replacing flat error strings. This establishes the error-reporting pattern for the sibling mojo-* parser libraries.

New module: src/xml/errors.mojo (exported from the package)

• line_col(source: Span[UInt8, _], offset: Int) -> Tuple[Int, Int] — 1-based line and column from a byte offset. Column is the 1-based byte offset within the line (no UTF-8 decode: cheap and deterministic even on invalid input). Clamps out-of-range offsets; only LF terminates a line, so CRLF contributes no phantom column.
• parse_error(msg: String, source: Span[UInt8, _], offset: Int) -> Error — message is exactly <msg> at line <L>, column <C>: '<snippet>', where the snippet is a ~30-byte whitespace-trimmed window of the offending line centered on the column, with ... where truncated, window edges nudged off UTF-8 continuation bytes, never multi-line.

Wiring

Every pull-parser raise site with a genuine byte offset now reports position + snippet: all strict-mode errors (previously a bare (line L, column C) suffix, no snippet) and the structural errors both modes raise (previously no position at all) — unterminated construct / start tag / attribute / attribute value, unquoted attribute value, malformed start/end tag, empty element name. Raise sites with no offset available (encoding sniff, entity-expansion guard, all DOM-level etree errors — events carry no offsets) are left untouched. Error mechanism unchanged: still raise Error(...), no new error types.

Verification

• 23 new tests in test/test_errors.mojo: line_col edge cases (offset 0, at-newline, past-end, empty source, CRLF, consecutive newlines), exact parse_error message formats including truncation and UTF-8-boundary behavior, and 3 hand-verified malformed-XML integration positions.
• Full suite green: 61 pull + 74 etree + 23 errors; CPython xml.etree anchor 14/14 matched; short fuzz run 0 crashes / 0 hangs; mojo format clean.

🤖 Generated with Claude Code