Call Gemini function even when text part is returned

[vlad-stoian]

Closes gh-2499

Sometimes the model returns a mix of text and function call parts:

content {
  role: "model"
  parts {
    text: "I will use the `get_current_weather` to get the current weather in San Francisco, Paris and Tokyo. I will set the unit to Celsius. Then I will print the temperature for each city.\n"
  }
  parts {
    function_call {
      name: "get_current_weather"
      args {
        fields {
          key: "location"
          value {
            string_value: "San Francisco, CA"
          }
        }
        fields {
          key: "unit"
          value {
            string_value: "C"
          }
        }
      }
    }
  }
  parts {
    function_call {
      name: "get_current_weather"
      args {
        fields {
          key: "location"
          value {
            string_value: "Paris, France"
          }
        }
        fields {
          key: "unit"
          value {
            string_value: "C"
          }
        }
      }
    }
  }
  parts {
    function_call {
      name: "get_current_weather"
      args {
        fields {
          key: "location"
          value {
            string_value: "Tokyo, Japan"
          }
        }
        fields {
          key: "unit"
          value {
            string_value: "C"
          }
        }
      }
    }
  }
}
finish_reason: STOP
avg_logprobs: -0.07267227420559177

An alternative implementation that I was thinking about was along the lines of:

		return candidate.getContent()
			.getPartsList()
			.stream()
			.filter(part -> part.hasText() || part.hasFunctionCall())
			.map(part -> {
				if (part.hasFunctionCall()) {
					FunctionCall functionCall = part.getFunctionCall();
					var functionName = functionCall.getName();
					var functionArguments = structToJson(functionCall.getArgs());
					var assistantToolCalls = List
						.of(new AssistantMessage.ToolCall("", "function", functionName, functionArguments));
					return new AssistantMessage("", messageMetadata, assistantToolCalls);
				}
				return new AssistantMessage(part.getText(), messageMetadata);
			})
			.map(message -> new Generation(message, chatGenerationMetadata))
			.toList();

But with the implementation above, only the first tool will be invoked, and then additional back and forth will happen between Spring AI and Gemini.

Please let me know if I there's anything I could improve.