[Bug]: Conversation restoration for legacy unphased stores fails to populate LLM context

Describe the bug

When restoring an agent conversation from an older, unphased store (e.g. legacy queen sessions), the messages are successfully read from disk (around core/framework/agent_loop/conversation.py:1783), but they are not properly re-written into the active LLM context.

As a result, when the agent resumes execution, it suffers from amnesia - behaving as if the previous turns never occurred because the LLM doesn't receive the history.

This is explicitly documented as a known bug and currently tracked via a skipped test in core/tests/test_event_loop_node.py:1092 (test_restore_legacy_unphased_assistant_message_preserves_store).

To Reproduce

1. Have a legacy unphased conversation store on disk (or run the skipped test test_restore_legacy_unphased_assistant_message_preserves_store).
2. Load the agent using Conversation.restore().
3. Notice that the actual prompt payload sent to the LLM does not contain the historical messages.

Expected behavior

Legacy unphased messages should be fully re-injected into the LLM's active prompt context during restore(), ensuring robust backward compatibility across Hive updates so user sessions are not broken.

Additional context

This significantly impacts backwards compatibility. Fixing this will allow us to un-skip the test in est_event_loop_node.py.