send_message/asend_message store Agent objects in memory content causing JSON serialization errors

[Harsh-617]

Describe the bug

LLMAgent.send_message and LLMAgent.asend_message store sender and recipients
as Python Agent objects inside the memory content dict. Because Agent objects are not
JSON serializable, this silently corrupts memory content and causes a hard crash whenever
recorded events are saved to JSON (e.g. via SimulationRecorder).

The built-in speak_to tool already stores unique_id values instead of objects, so
the two messaging mechanisms are currently inconsistent.

To Reproduce

# Crash via SimulationRecorder
@record_model
class MyModel(mesa.Model):
    ...

model = MyModel()
agent_a.send_message("hello", [agent_b])
model.save_recording()          # TypeError: Object of type LLMAgent is not JSON serializable

# Or directly:
import json
agent_a.send_message("hello", [agent_b])
json.dumps(agent_a.memory.step_content)  # same error

Expected behavior

Message memory entries should store serializable identifiers:

content={
    "message": message,
    "sender": self.unique_id,
    "recipients": [r.unique_id for r in recipients],
}

This is already what speak_to does (llm_agent.py lines 213–218) and what
serialization-safe memory handling requires.

Current behavior

Both send_message (line 284) and asend_message (line 269) in llm_agent.py store:

content={
    "message": message,
    "sender": self,           # Agent object — not serializable
    "recipients": recipients, # list of Agent objects — not serializable
}

Additional context

Affects: llm_agent.py lines 269–273 (asend_message) and 284–288 (send_message)
The bug is silent until JSON serialization is attempted, making it hard to catch without recording enabled
speak_to (inbuilt_tools.py lines 213–218) handles this correctly and should be the reference implementation