[ryanmurf]

Problem

rtk read <file> --tail-lines N -n renumbered the surviving tail window starting from 1. A model reading the output was told the last N lines of the file were lines 1..N, when they are actually the final N lines. Example (5-line file):

$ rtk read foo.txt --tail-lines 2 -n
1 │ l4      <- WRONG: this is line 4
2 │ l5      <- WRONG: this is line 5

This is the output-mangling class (same family as the grep #6 / wc #8 / tree #9 / find #10 fixes): a filter corrupts the information the model reasons over — here, the line references it would cite.

Fix

• apply_line_window now returns (windowed_text, start_offset) where start_offset is the 1-based original position of the first emitted line (total - tail + 1 for a tail window, 1 otherwise).
• format_with_line_numbers renders the gutter from that offset; gutter width is sized for the largest number actually printed so a high offset (e.g. lines 9–10) stays aligned.
• --max-lines/smart_truncate keeps a non-contiguous selection starting at line 1, so it correctly stays 1-based (already signals omissions via the [N more lines] marker).

After:

$ rtk read foo.txt --tail-lines 2 -n
4 │ l4
5 │ l5

Tests

Adds a RED-on-revert regression (test_format_line_numbers_with_tail_offset — fails with 1 │ l4 if the offset wiring is reverted) plus edge cases: width across a digit boundary, empty content (no panic), tail=0, tail>len, and the no-window / max-lines offset invariants.

Verification

• cargo fmt --check clean
• cargo clippy --all-targets -- -D unsafe_code clean
• cargo test --all green (1752 + 12 pass)
• cargo build --release + manual E2E confirm the corrected gutter

[github-actions]

👋 Thanks for the PR! It looks like this targets master, but all PRs should target the develop branch.

Please update the base branch:

1. Click Edit at the top right of this PR
2. Change the base branch from master to develop

See CONTRIBUTING.md for details.