Parse errors carry line/column + context snippet

[conorbronsdon]

Adopts the suite-wide error-reporting pattern from conorbronsdon/mojo-xml#3: parse errors now carry a 1-based line/column position and a snippet of the offending line.

What changed

• New html.errors module — a verbatim copy of mojo-xml's errors.mojo. line_col(source, offset) maps a byte offset to a 1-based (line, column) pair (column = byte offset within the line; LF-only terminator, CRLF-safe), and parse_error(msg, source, offset) builds an Error reading <msg> at line <L>, column <C>: '<snippet>' — snippet is the whitespace-trimmed offending line, windowed to ~30 bytes centered on the offset, ... on cut sides, never split mid-UTF-8-sequence, never multi-line. Both exported from the package.
• All strict-mode tokenizer errors wired through it. mojo-html's liberal mode never raises on malformed markup by design; every raise site lives behind strict=True and already had a byte offset, so all of them now carry position + snippet (previously a bare (line L, column C) suffix with no snippet):
  • mismatched end tag / stray end tag — at the offending end tag
  • unclosed element at EOF — now points at the unclosed start tag instead of the useless EOF position (new _open_pos bookkeeping, strict mode only)
  • unterminated comment / CDATA — at the construct opener
  • unknown entity, bare &, malformed numeric reference — now point at the offending & (in text, attribute values, and escapable raw text) instead of the tokenizer's position after the run
• Mechanism unchanged: still a plain raise Error(...); the repo-specific mojo-html [strict]: prefix stays in the caller; existing contains=-style assertions keep matching.
• Tests: new test/test_errors.mojo (31 tests) — the full line_col/parse_error edge-case suite ported from mojo-xml, plus 11 integration tests asserting hand-verified line/column/snippet output for every wired raise site. Registered in the pixi test task and the CI workflow. Full suite: 99 tests green.

🤖 Generated with Claude Code