feat: add opt-in conversation checkpoint and resume

[bhargavikalicheti]

Summary

Add an initial checkpoint/resume abstraction for #2340. Applications can save explicitly serializable conversation and workflow state, then restore it into a new session after an interruption.

• Add versioned checkpoints and participant snapshot, validation, and restore hooks.
• Provide pluggable storage with an in-memory implementation.
• Require application-defined resume policies and enforce expiration.
• Validate checkpoint identity, participant sets, and state versions before restoring.
• Include a SQLite example demonstrating recovery across separate processes.
• Add documentation and a Windows/Linux CI workflow.

Scope and limitations

This is an opt-in proof of concept pending maintainer feedback. It does not automatically reconnect transports or checkpoint runtime resources. Applications remain responsible for authorization, consistent snapshot boundaries, and tool-operation idempotency.

Restore hooks are not transactional. If a hook fails after another participant has restored, the application must discard the partially restored session.

Validation

• All 21 tests pass locally, including cross-process recovery.
• Formatting and lint checks pass.
• Package builds and installs successfully; tests also pass against the installed copy.

Refs #2340

[bhargavikalicheti]

@plutoless @halajohn Hi, I’ve prepared an initial proof of concept for this issue with opt-in state hooks, versioned checkpoints, pluggable storage, expiration, and application-defined resume policies.
It includes documentation, a SQLite example demonstrating recovery across separate processes, and 21 passing tests. Restore is explicit; automatic transport reconnection and distributed failover remain outside scope.
Before expanding the implementation, I’d appreciate feedback on whether a standalone Python system package is the right home for checkpoint coordination, and whether the proposed participant hooks fit TEN’s architecture.