Fix dspy task 8563: update TOOL_CALL_TEST_CASES in features 2-6 test patches

[AlienKevin]

Summary

• The combined.patch changes ToolCalls.format() return type from list[dict] to dict when no metadata is present
• Feature 1's test patch correctly updates TOOL_CALL_TEST_CASES expectations and test_tool_calls_format_from_dict_list assertions to match the new dict format
• Features 2–6 test patches do not include this update
• Since each feature's tests run against the full merged implementation (via runner.sh tests_patch merged.patch), the pre-existing test_tool_calls_format_basic[tool_calls_data0-expected0] fails whenever features 2–6 tests run — the test still expects the old list format but format() now returns a dict

Fix

Add the following updates from feature 1's test patch to features 2–6:

1. Import update: add convert_input_schema_to_tool_args to the import line
2. TOOL_CALL_TEST_CASES data update: [{"type": "tool_calls", ...}] → {"tool_calls": [...]}
3. test_tool_calls_format_from_dict_list assertion update: result[0]["tool_calls"] → result["tool_calls"]

Test plan

• Oracle test passes for all dspy task 8563 feature pairs (including f2-4 which previously failed)
• Verified on Modal with Harbor evaluation framework