Fix repeated Deep Agents tool calls with identical arguments

[1fanwang]

What was changed

When a Deep Agent requested the same tool and arguments twice in one workflow,
only the first call ran. The second call returned the first result without
scheduling its Temporal Activity.

run_deep_agent caches activity results for continue-as-new. The tool cache key
now includes the call ID as well as the tool name and arguments. Separate calls
run independently, while replay of the same call can still reuse its result.

Why?

Before this fix, a side-effecting tool requested twice with identical arguments
ran only once. After the fix, workflow history records two
deepagents.invoke_tool Activity schedules.

Checklist

1. Closes

No issue. This is a small, self-contained bug fix.

2. How was this tested:

The regression test drives a real Deep Agents workflow against a local Temporal
server and makes two tool calls with the same arguments but different call IDs.

Red on main

$ git checkout origin/main -- temporalio/contrib/deepagents/workflow.py
$ uv run pytest tests/contrib/deepagents/test_tools.py::test_repeated_tool_calls_with_same_args_both_run -q --no-header --tb=short -p no:pretty
E   assert 1 == 2
FAILED tests/contrib/deepagents/test_tools.py::test_repeated_tool_calls_with_same_args_both_run
1 failed, 18 warnings in 3.58s

Green with this fix

$ git checkout HEAD -- temporalio/contrib/deepagents/workflow.py
$ uv run pytest tests/contrib/deepagents/test_tools.py::test_repeated_tool_calls_with_same_args_both_run -q --no-header --tb=short -p no:pretty
.                                                                        [100%]

3. Any docs updates needed?

The changelog includes the corrected behavior. The public API is unchanged.