[checkpoint] Restore conversational tool-use simulation trajectories

Parent

Child of #3219 and #3024.

Problem and state owner

resources_servers/conversational_tool_use_simulation/app.py owns the evolving conversation, current actor, pending tool calls, artifacts, failures, terminal state, and judge results in ConversationSessionState.

Proposed adapter

Version and serialize the complete logical trajectory, including phase/actor and pending-call disposition; do not re-run simulator generations already represented in the snapshot.

Implement explicit checkpoint opt-in, route classification, export/restore/retire hooks, execution-to-session binding, revision reporting, atomic replacement under attempt N+1, and stale-attempt fencing.

Acceptance criteria

• Restore after user, assistant, and pending-tool boundaries selects the exact next actor and never repeats a committed simulator/model operation.
• Checkpoint after multiple mutations, replace the server process, restore, and compare the next operation with an uninterrupted execution.
• Corrupt or incomplete bulk state activates no sessions.
• Attempt N cannot mutate or retire restored attempt N+1.
• Terminal cleanup is idempotent and releases restored state exactly once.

Start with resources_servers/conversational_tool_use_simulation/app.py and inspect ConversationSessionState and the existing conversation execution flow. Trace how actors, pending tool calls, artifacts, failures, terminal state, and judge results are mutated; done means the acceptance scenarios pass, including exact next-actor recovery, corrupt-state rejection, stale-attempt fencing, and idempotent cleanup.