Fix: handle JSON parse error in streaming tool calls

[kmw10693]

Overview

This PR fixes a JSON parsing error that occurs when handling streaming tool calls in Spring AI.

Related issue: spring-projects/spring-ai#4250

---

Why

• In streaming mode, tool call JSON fragments can arrive incomplete.
• This results in Jackson throwing errors such as: com.fasterxml.jackson.core.JsonParseException: Unexpected close marker ']': expected '}'
• The issue only occurs in streaming (stream()) calls.
• Synchronous calls (call()) are not affected.
• This prevents users from reliably using tool callbacks with streaming responses.

---

What Changed

• Added a new ToolExecutionEligibilityPredicate implementation: ToolOnFinishPredicate.
• Ensures tool execution only runs after the assistant signals completion (finishReason = "tool_calls" or "stop").
• Prevents premature execution and malformed JSON during streaming tool calls.

---

How to Use

Example usage:

ChatClient client = ChatClient.create(openAiChatModel);

String result = client.prompt()
  .user("Migrate node cluster xx")
  .toolCallbacks(ToolCallbacks.from(new MyTool()))
  .options(OpenAiChatOptions.builder()
      .toolExecutionEligibilityPredicate(new ToolOnFinishPredicate())
      .build())
  .stream()
  .content();

[markpollack]

can't use this annotation for not null

[kmw10693]

@markpollack I’ve removed ToolOnFinishPredicate.java and applied the fix directly in OpenAiChatModel.
when the LLM returns malformed or invalid JSON during streaming, the code explicitly throws an exception.

[markpollack]

The PR looks incomplete. I've commented on the issue thread for further clarification. #4274

[kmw10693]

The PR looks incomplete. I've commented on the issue thread for further clarification. #4274

@markpollack
On the issue, the problematic LLM was Qwen3-8B. they found that the chunk returned by the LLM was invalid.

However, the OpenAIChatModel was not properly detected when the LLM returned an invalid response.
I’ve updated OpenAiChatModel so that if the LLM sends an invalid or malformed JSON response (e.g., broken chunk). the code now explicitly throws an exception. I also added the OpenAiChatModelStreamingTest