fix: respect user-specified git log limits (#461)

[ousamabenyounes]

Summary

Fixes #461 — rtk git log --oneline silently truncates output to 10 entries regardless of user intent.

Root cause: Two layers of truncation were fighting each other:

1. run_log() injected -10 when no -N flag was present (even with --oneline)
2. filter_log_output() applied .take(limit) again, re-capping even when the user explicitly passed -N
3. Lines >80 chars were truncated, hiding PR numbers and full author names

Why this matters for LLM workflows:

• git log --oneline -30 for rebase/squash: only got 10 commits, forcing re-runs
• git log --oneline for changelog: silently incomplete history
• PR numbers like (#473) and author names were truncated at 80 chars

Changes

Scenario	Before	After
rtk git log --oneline (no -N)	10 entries	50 entries
rtk git log -20	20 entries (but re-truncated)	20 entries (no re-truncation)
rtk git log (no flags)	10 entries	10 entries (unchanged)
Line truncation with explicit -N	80 chars	120 chars

Before/After

# BEFORE: --oneline silently capped at 10
$ rtk git log --oneline | wc -l
10

# AFTER: --oneline returns 50 (sensible default for compact format)
$ rtk git log --oneline | wc -l
50

# BEFORE: -20 truncated lines at 80 chars
cc93afc chore(master): release 0.28.2 (#473) (17 hours ago) <github-actions[b...

# AFTER: -20 uses 120-char threshold (preserves PR numbers + authors)
cc93afc chore(master): release 0.28.2 (#473) (17 hours ago) <github-actions[bot]>

Test plan

• 2 new unit tests: test_filter_log_output_user_limit_no_cap, test_filter_log_output_user_limit_wider_truncation
• All 7 filter_log_output tests pass
• Full suite: 766 passed, 0 failed
• Manual verification: --oneline, -20, default (no flags)
• cargo fmt && cargo clippy && cargo test all green

Generated with Claude Code

[pszymkowiak]

Thanks @ousamabenyounes! The user_set_limit approach is clean and the unicode fix with chars().count() is a real improvement.

I rebased onto develop and resolved the conflict (the limit logic had evolved there). I also pushed two fixes:

1. Bug: -n 20 and --max-count=20 extraction — has_limit_flag correctly detected these forms, but the extraction block only matched the combined -20 form. Extracted a parse_user_limit() function that handles all 4 forms: -20, -n 20, --max-count=20, --max-count 20.

2. Token savings test — Added test_filter_log_output_token_savings asserting ≥60% savings per our testing policy.

799 tests pass. Merging into develop.