perf(visualizer): fix O(n²) chars().nth scan and reduce allocations

[Boshen]

Summary

• Bug fix in generate_line_utf16_tables: content.chars().nth(i + 1) treated the byte offset i (from char_indices()) as a char index. That made the \r\n check walk the string from the start on every line break (O(n²) over the loop) and return the wrong char whenever earlier content contained multi-byte UTF-8. \n is always a single ASCII byte, so peeking content.as_bytes().get(i + 1) is both O(1) and correct.
• Replace s.push_str(&format!(...)) in SourcemapVisualizer::get_text with writeln!(s, ...) — formats directly into the output String instead of allocating an intermediate one per token.
• Mark trivial Token accessors (get_dst_line, get_dst_col, get_src_line, get_src_col, get_source_id, get_name_id) #[inline]. Release+LTO already inlines them through serialize_mappings, but the annotation helps debug builds and downstream crates that don't enable LTO.

No behavior change — visualizer insta snapshots match exactly.

[codspeed-hq]

Merging this PR will not alter performance

✅ 16 untouched benchmarks
⏩ 5 skipped benchmarks¹

---

_{Comparing perf/visualizer-and-token-accessors (6d20ba6) with main (55ebb8e)}

Footnotes

1. 5 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩