ApigeeLlm: tool call arguments that are valid JSON but not an object crash the non-streamed path and pass through as args when streamed #7740

🔴 Required Information

Describe the Bug:
When a Chat Completions tool call's arguments is complete JSON but not an object ([1, 2], "x", 5, null), ChatCompletionsResponseHandler handles it two different ways:

• Non-streamed (process_response): _upsert_tool_call parses the string and then calls function_call.args = dict(args) (apigee_llm.py:1370 on main), which raises TypeError or ValueError. The response never reaches the caller.
• Streamed (process_chunk): _get_content_parts assigns the parsed value directly (apigee_llm.py:1285), so FunctionCall.args becomes a list, string, int or None and the call goes on to tool execution.

This is independent of the finish reason (same with tool_calls and stop) and separate from the truncated-arguments case in #7648 / #7686, where it came up (surajksharma07 spotted it there).

Steps to Reproduce:

1. Use current main at bf6739723 (package version 2.13.0).
2. Run the script below. It feeds synthetic responses to the handler; no credentials or network.

from google.adk.models.apigee_llm import ChatCompletionsResponseHandler

def nonstream(args, finish):
  resp = {"model": "m", "choices": [{"finish_reason": finish, "message": {"role": "assistant", "content": None,
          "tool_calls": [{"id": "c1", "type": "function", "function": {"name": "w", "arguments": args}}]}}]}
  try:
    r = ChatCompletionsResponseHandler().process_response(resp)
    return f"args={[p.function_call.args for p in r.content.parts if p.function_call]}"
  except Exception as e:
    return f"{type(e).__name__}: {e}"

def stream(args, finish):
  h = ChatCompletionsResponseHandler()
  out = list(h.process_chunk({"model": "m", "choices": [{"delta": {"role": "assistant", "tool_calls": [
      {"index": 0, "id": "c1", "type": "function", "function": {"name": "w", "arguments": args}}]}, "finish_reason": None}]}))
  out += list(h.process_chunk({"model": "m", "choices": [{"delta": {}, "finish_reason": finish}]}))
  final = [r for r in out if not r.partial]
  return f"args={[p.function_call.args for r in final for p in r.content.parts if p.function_call]}"

for a in ["[1, 2]", '"x"', "5", "null"]:
  for f in ["tool_calls", "stop"]:
    print(f"{a:8} {f:10} non-stream: {nonstream(a, f)}")
    print(f"{a:8} {f:10} stream:     {stream(a, f)}")

Expected Behavior:
Both paths treat non-object arguments the same way, and neither hands a non-dict FunctionCall.args to tool execution. For example: reject the call with a clear error (or a function-response error the model can recover from), consistently on both paths.

Observed Behavior:

Same output with finish_reason tool_calls and stop.

Environment Details:

• ADK: main bf6739723 (2.13.0)
• Python: 3.12.13
• OS: macOS

Model Information:

• Any model behind ApigeeLlm using the Chat Completions API. Reproduced with synthetic responses, no live call.

🟡 Optional Information

Additional Context:
n1shanthb offered to take the fix in #7648:
@n1shanthb