[Bug]: Glob can overflow agent context with unbounded output

### Describe the bug

The built-in `Glob` tool currently joins every matching file path into one tool result. For a broad pattern such as `**/*`, the result can grow with the entire workspace and add an unnecessarily large message to the agent context.

The built-in `Grep` tool already has a default `head_limit` of 250 entries and supports `offset`, but `Glob` has no equivalent default result limit or pagination mechanism.

This is not an exception. The problem is that the tool has no default result limit and no pagination mechanism:

```text
broad Glob pattern
|
+-- match 1
+-- match 2
+-- ...
+-- match 201
|
+-- current: all paths are returned in one result
+-- expected: a bounded page with a way to fetch more
```

### To reproduce

1. Create more than 200 files matching the same pattern.
2. Call the built-in `Glob` tool without an explicit result limit:

```python
result = await Glob()(pattern="match_*.log", path="/tmp/workspace")
```

3. Observe that every matching path is returned in one result and no truncation guidance is provided.

Example observation with 201 matching files:

```text
matched_paths=201
truncation_notice=False
```

No exception is raised.

### Expected behavior

`Glob` output should remain bounded even when the caller omits optional arguments:

- enforce a tool-side default result limit;
- allow callers to request a smaller or larger page within a hard cap;
- include an explicit truncation notice when more matches exist;
- provide a way to request later pages.

### Additional context

The proposed defaults for `Glob` are a default limit of 200 paths and a hard maximum of 1000 paths. The proposed first version applies the limit at the tool-output boundary; moving the limit into helper traversal can be handled as a separate optimization.