[fix] parse_tool_calls: replace shared dict refs from list multiplication with independent dicts

[NIK-TIGER-BILL]

Problem

parse_tool_calls in libs/agno/agno/models/openai/chat.py uses:

tool_calls.extend([{}] * (_index - len(tool_calls) + 1))

This is a classic Python pitfall — [{}] * N creates N references to the same dict object, not N independent dicts. When any element is mutated (e.g. tool_call_entry["id"] = ...), all N elements change simultaneously through the shared reference.

This causes every tool call to be executed twice (or more) when the model returns a non-zero-based tool_call.index (e.g., Claude via OpenRouter), because multiple slots in the list end up sharing the same dict and all get updated together.

Fix

Replace [{}] * N with a list comprehension:

# Before (buggy)
tool_calls.extend([{}] * (_index - len(tool_calls) + 1))

# After (fixed)
tool_calls.extend([{} for _ in range(_index - len(tool_calls) + 1)])

Each slot now gets its own independent empty dict, so mutations to one slot do not affect others.

Verification

# Demonstrates the bug
shared = [{}] * 3
shared[0]['key'] = 'value'
assert shared[1]['key'] == 'value'  # all 3 are mutated!

# Correct
independent = [{} for _ in range(3)]
independent[0]['key'] = 'value'
assert 'key' not in independent[1]  # only index 0 is mutated ✅

Closes #6542