0.2.0 — Mojo 1.0

• Builds on released Mojo 1.0. The pixi channel moved from
  max-nightly to max and the dependency pin from >=1.0.0b3.dev… to
  >=1.0,<2; CI installs mojo==1.0.0 from PyPI instead of the nightly
  index with --prerelease allow. The weekly drift check still runs
  against the latest nightly, so the repo still reports when the language
  moves under it.

• 1.0 source fix. The positive/negative keyword lists in
  main_text_confident are now annotated List[String]. Under 1.0 a
  bare list expression builds an Array, which no longer converts to
  _class_weight's List[String] parameter.

• README test count corrected: 63 → 99. Robustness re-checked with 450
  mutated documents — zero crashes, zero hangs.

• New html.errors module (exported from the package), following the
  shared mojo-* suite pattern: line_col(source, offset) maps a byte
  offset to a 1-based (line, column) pair — the column is the 1-based
  BYTE offset within the line, no UTF-8 decoding — and
  parse_error(msg, source, offset) builds an Error reading
  <msg> at line <L>, column <C>: '<snippet>', where the snippet is up
  to ~30 bytes of the offending line centered on the column,
  whitespace-trimmed, with ... where truncated, and never multi-line.

• Strict-mode tokenizer errors now carry that position + snippet
  (previously a bare (line L, column C) suffix with no snippet), with
  more useful offsets: unclosed-element-at-EOF errors point at the
  unclosed start tag instead of the end of input, and entity errors
  (unknown entity, bare &, malformed numeric reference) point at the
  offending & — in text, attribute values, and escapable raw text —
  instead of the tokenizer's position after the run.

• No mechanism change: the tokenizer still raises a plain Error, the
  mojo-html [strict]: prefix is unchanged, and existing
  contains=-style message checks keep matching. The default liberal
  mode still never raises on malformed markup.