fix(positions): remaining column inconsistencies after #797 #830

Follow-up items from the Opus review of PR #826 (refs #797).

1. Quoted identifier newline error uses rune-based column instead of byte column

pkg/sql/tokenizer/tokenizer.go, readQuotedIdentifierSlow (~line 1318): the newline-terminated quoted identifier error builds its models.Location from the rune-based internal Position.Column instead of t.toSQLPosition(startPos) (byte-based). This contradicts the models.Location doc, which specifies byte columns.

The EOF branch already uses t.toSQLPosition(startPos).

Repro: x\t/* c */ é\t"ab\nc" reports column 13, but the opening quote is at byte column 14.

The same pattern exists in the unused StringLiteralReader (string_literal.go).

Fix:

• Use t.toSQLPosition(startPos) in the newline branch, matching the EOF branch.
• Apply the same fix in string_literal.go.
• Add a multi-byte regression test that asserts the byte column.

2. L006 column alignment reports the wrong column

pkg/linter/rules/style/column_alignment.go: getIndentSize counts a tab as 4 columns and the rule reports Location.Column = indent+1. The reported column should be a byte offset.

Fix:

• Keep tab = 4 when comparing indentation widths.
• Report the byte offset of the first non-whitespace character, plus 1, as Location.Column.

3. Tighten TestTokenPositions_CommentsKeepOwnPositions

Strengthen the test to assert tokens[2].Start >= Comments[0].End, so it checks that tokens after a comment do not overlap the comment's span.

🤖 Generated with Claude Code

https://claude.ai/code/session_01RkzVu3JY9Y6BRWfjesWts9