Describe the bug

speak_to call crashes with AttributeError when an
LLM agent calls it and any of the named recipients is a plain (non-LLM) Mesa agent.

The tool looks up recipients by scanning agent.model.agents which includes all
agents and not just LLMAgents, then unconditionally calls
recipient.memory.add_to_memory(). Plain Mesa agents have no memory attribute,
so this raises:

AttributeError: 'RuleBasedAgent' object has no attribute 'memory'

Expected behavior

When a recipient has no memory attribute, speak_to should skip that recipient silently (not crash the simulation) and emit a logger.warning so the developer can see which agent was skipped

LLM recipients in the same call should still receive the message normally.

To Reproduce

import os
os.environ["GEMINI_API_KEY"] = "just_testing"

from mesa.model import Model
from mesa.agent import Agent
from mesa.space import MultiGrid
from unittest.mock import MagicMock
from mesa_llm.llm_agent import LLMAgent
from mesa_llm.reasoning.cot import CoTReasoning
from mesa_llm.tools.inbuilt_tools import speak_to

class RuleBasedAgent(Agent):
    def step(self): pass

class MixedModel(Model):
    def __init__(self):
        super().__init__(seed=42)
        self.grid = MultiGrid(5, 5, torus=False)

model = MixedModel()
llm_agent = LLMAgent(model=model, reasoning=CoTReasoning)
rule_agent = RuleBasedAgent(model=model)
model.grid.place_agent(llm_agent, (1, 1))
model.grid.place_agent(rule_agent, (2, 2))
llm_agent.memory.add_to_memory = MagicMock()

speak_to(llm_agent, [rule_agent.unique_id], "Hello!") #will crash here

Additional context

• The existing test test_speak_to_records_on_recipients manually attaches a mock .memory to every test recipient, which is why it passes even on the broken code
• Mixed-agent models (LLM agents + rule-based agents) are a standard Mesa pattern; this bug makes cross-type communication difficult