[psteinroe]

Stacked on #1063.

The suppressions parser numbers lines with str::lines, which ends lines at \n (and drops the \r of a \r\n), but looks up their offsets and the lines of diagnostics in its own line index, which also ended a line at a lone \r. After a lone \r, the two disagree, and a line suppression points at the wrong text and doesn't apply to the line below it:

select '\r';
-- pgls-ignore lint
select 1; -- still reported

The parser now uses pgls_line_index::LineIndex, which ends lines at \n only, like str::lines, the language server and the WASM bindings. Its own line_index.rs is gone. LineIndex gets a Default (the index of an empty text), since Suppressions derives it.

This changes behaviour only for files with a lone \r: lines after it are numbered one lower than before, which is what editors that end lines at \n show too. \r\n and \n files are numbered as before.

Tests: a_lone_carriage_return_does_not_end_a_line checks that the suppression above applies and covers the comment's text; it fails without this change. line_breaks_and_wide_characters in pgls_line_index (#1063) pins the index's line numbers for \r\n, a lone \r and multi-byte characters. cargo test -p pgls_suppressions -p pgls_line_index, cargo clippy --workspace --all-targets -- -D warnings.

Merging with #1008: that stack also edits crates/pgls_suppressions/src/line_index.rs (line_for_offset with partition_point). Whichever merges second gets a modify/delete conflict there; keep the deletion, since the shared index already looks lines up with partition_point. parser.rs merges cleanly, and the Suppressions::default() it returns early works with the new Default.