Widen RunHooks.on_tool_end result type to Any

[NishchayMahor]

Closes #3512.

Summary

Tools can return any JSON-serialisable value — strings, Pydantic models, dicts, lists. The runtime forwards whatever the tool returned (after output guardrails) straight into `hooks.on_tool_end`:

```python

src/agents/run_internal/tool_execution.py

final_result = await _execute_tool_output_guardrails(...) # -> Any
self.hooks.on_tool_end(tool_context, self.public_agent, func_tool, final_result)
```

But `RunHooksBase.on_tool_end` and `AgentHooksBase.on_tool_end` declared `result: str`. With strict type checking that triggers:

```
TypeError: Argument 'result' has incorrect type (expected str, got dict)
```

The fix is to align the hook signature with what the runtime actually passes — widen `result` to `Any` and document what the parameter contains.

What changed

• `src/agents/lifecycle.py` — `RunHooksBase.on_tool_end` and `AgentHooksBase.on_tool_end` now take `result: Any`, with a docstring note that this is the raw post-guardrail value the tool returned.
• `tests/test_run_hooks.py` — new `test_on_tool_end_receives_structured_dict_result` regression that runs a `function_tool` returning `dict[str, Any]` and asserts the hook sees the raw dict.

No call sites changed shape — they were already passing through the raw value.

Compatibility

Backward-compatible widening. Existing overrides that typed the parameter `str` keep working because `str` is a subtype of `Any`; the only behavioural change is that strict runtime type checks now accept the dict/list/Pydantic-model values the runtime was already passing in.

Verification

```
$ uv run pytest tests/test_run_hooks.py tests/test_global_hooks.py tests/test_agent_llm_hooks.py -q
21 passed in 0.18s

$ uv run ruff check src/agents/lifecycle.py tests/test_run_hooks.py
All checks passed!

$ uv run pyright src/agents/lifecycle.py tests/test_run_hooks.py
0 errors, 0 warnings, 0 informations
```

[seratch]

Thanks for sharing this. Resolved by #3518