fix(read): don't count --max-lines truncation as token savings (#1045)

[ousamabenyounes]

Summary

• rtk read file --max-lines N was comparing the full raw file against the truncated output in timer.track(), inflating reported savings by the entire skipped portion — 250M+ tokens reported for a single command on large files
• Fix: capture the post-filter, pre-truncation content and use it as the tracking baseline when --max-lines or --tail-lines is active
• Same fix applied to run_stdin()

Root cause

// BEFORE — bug
timer.track("cat file", "rtk read", &content, &rtk_output);
//                                   ^^^^^^^^   ^^^^^^^^^^
//                                   11MB file  5 lines → fake 2.8M token "savings"

// AFTER — fix
let tracking_output = filtered_for_tracking.as_deref().unwrap_or(&rtk_output);
timer.track("cat file", "rtk read", &content, tracking_output);
//                                             ^^^^^^^^^^^^^^
//                                             pre-truncation filtered content → real savings only

Test plan

4 new regression tests in src/cmds/system/read.rs:

• test_max_lines_tracking_uses_pretruncation_content — tracking baseline equals full filtered content, not truncated slice
• test_tail_lines_tracking_uses_pretruncation_content — same invariant for --tail-lines
• test_no_truncation_tracking_unchanged — without truncation flags, behaviour is identical to before
• test_savings_inflation_is_zero_with_fix — numerical proof: buggy code reports >1k fake tokens, fix reports 0

cargo test: 1347 passed, 0 failed

Fixes #1045

🤖 Generated with Claude Code