fix(tools/glob): don't report truncation at exactly 100 matches in ripgrep backend

[tomsen-ai]

HUMAN:

Found that the glob tool reports results as truncated when there are

exactly 100 matches, even though nothing was actually dropped. The
ripgrep and Python backends were inconsistent on this boundary, and the false flag misleads the agent into doing extra re-searches. I went
through the root cause and the before/after repro myself.

---

AGENT:

Why

The glob tool's ripgrep backend stops collecting at 100 files and then flags truncated = len(file_paths) >= 100. A result set of exactly 100 files is therefore reported as truncated even though nothing was dropped:

• impl.py Python fallback uses a strict > 100 (_execute_with_glob), and the grep tool does too — the ripgrep path is the only >=.
• definition.py's field doc says "Whether results were truncated to 100 files" — false at exactly 100.
• The flag is not cosmetic: when set, the observation appends "[Results truncated to first 100 files. Consider using a more specific pattern.]", instructing the agent to re-search for files that don't exist.

Because the backend is chosen by whether rg is installed, the same directory + pattern currently yields different truncated answers on different machines.

Summary

• Collect all ripgrep stdout lines first (already sorted by --sortr=modified), set truncated = len(lines) > 100, return the first 100 — matching the Python fallback exactly.
• Path.resolve() now runs only on the returned slice instead of every line.
• Add a red→green boundary test asserting both backends return len == 100 and truncated is False at exactly 100 matches.

Issue Number

N/A (searched open/closed issues and PRs for glob truncation — no prior report).

How to Test

uv run pytest tests/tools/glob/ -q

The new test fails on main (ripgrep backend reports truncated=True at 100 matches) and passes with this change. End-to-end repro against the real executor (ripgrep installed):

# main
matches=100: returned=100 truncated=True   ← false positive
# this branch
matches=99:  returned=99  truncated=False
matches=100: returned=100 truncated=False
matches=101: returned=100 truncated=True

Video/Screenshots

Agent-visible observation text on main at exactly 100 matches (real GlobExecutor run, not mocked):

Found 100 file(s) matching pattern '*.py' in '/tmp/...':
...
[Results truncated to first 100 files. Consider using a more specific pattern.]

After this change the misleading instruction no longer appears; the 101-match case still produces it correctly.