fix(git): correct git_log off-by-one in timestamp-filtered output

[ly-wang19]

Description

git_log's timestamp-filtered branch uses the pretty-format %H%n%an%n%ad%n%s%n — note the trailing %n after the subject. With git's inter-commit newline, .split('\n') then yields 5 fields per commit (hash, author, date, subject, ""), but the parser strides in groups of 4 (range(0, len(log_output), 4)).

So only the first commit parses correctly; every subsequent commit is shifted by one — Author shows the next commit's hash, Date shows its author, etc. — producing scrambled git_log output whenever start_timestamp / end_timestamp is used on a repo with 2+ matching commits.

The sibling no-timestamp branch (which uses iter_commits) parses cleanly, establishing the intended one-record-per-commit contract that this branch violates.

Fix

Drop the trailing %n (--format=%H%n%an%n%ad%n%s) so each commit is exactly four newline-delimited fields, matching the groups-of-4 parser.

Testing

Added test_git_log_with_timestamp_filter_parses_each_commit (three commits + a start_timestamp filter): before the fix it fails (messages/authors are shifted into each other); after the fix the three commits parse in order. Full src/git suite: 44 passed.