Parse errors carry line/column + context snippet

[conorbronsdon]

Adds position info to every parse-time error, following the shared errors.mojo pattern established in conorbronsdon/mojo-xml#3.

• New template.errors module, copied verbatim from mojo-xml: line_col(source, offset) (1-based line; column = 1-based byte offset in the line, LF-only terminators, clamped) and parse_error(msg, source, offset) producing <msg> at line <L>, column <C>: '<snippet>' with a ~30-byte, UTF-8-safe, single-line snippet of the offending line. Both exported from the package.
• Lexer Tokens carry byte offsets (tag opener + inner-content start) instead of line numbers; expression tokens carry absolute byte offsets. Every lexer/parser raise site now reports an exact position: unclosed tags and unclosed {% if %} / {% for %} point at the opening tag (not EOF), expression errors (unexpected character, unterminated string literal, trailing tokens, missing ]/)/names, depth-cap guards) point at the offending byte. Guard conditions untouched — only error construction changed.
• Render-time errors (undefined variables, eval-depth cap) are unchanged: the AST carries no source offsets, and threading them through was out of scope.
• 26 new tests in test/test_errors.mojo (ported edge-case suite + 6 integration tests with hand-verified line/column), wired into the pixi test task and CI as a separate step. Full suite: 104 tests green; mojo format clean.

🤖 Generated with Claude Code