[Feature] Add built-in session inspection to the T3 CLI and agent toolset

[Albro3459]

I'd love to see the functionality provided by t3-session become a first-class part of T3 Code.

t3-session provides an agent-first read-only CLI/API for inspecting T3 Code's persisted conversation data, and I've found it extremely useful for scripting, debugging, automation, and giving other tools access to existing T3 sessions without requiring them to understand T3's underlying storage format.

Some of the particularly useful capabilities are:

• Listing persisted threads, with filtering and pagination
• Finding/retrieving threads and their conversation history
• Retrieving only the latest turn or a bounded window of turns
• Machine-readable JSON/JSONL output
• Inspecting whether a thread appears active or complete
• Tailing a live thread and receiving incremental updates
• Inspecting participants/subtasks and their hierarchy

For example, having official CLI commands along the lines of:

t3 sessions list
t3 sessions get <thread-id>
t3 sessions get <thread-id> --last-turn --format json
t3 sessions tail <thread-id>
t3 sessions participants <thread-id> --tree

would make T3 Code much easier to integrate into scripts, editor tooling, automation, and other developer workflows.

Agent tooling

I think this could be especially useful as a T3-specific tool/skill available to agents, not just as a user-facing CLI.

I know T3 Code isn't necessarily trying to be a general-purpose agent harness, but session inspection feels like functionality that is uniquely useful inside T3 itself. Giving agents a supported way to query T3's own session state could make them much more effective at working within the application.

For example, an agent could:

• Look up another or previous T3 thread when the user references earlier work
• Pull only the latest few turns instead of loading an entire conversation
• Check whether another thread or task is still active
• Inspect the participants/subagents involved in a thread and their current state
• Follow another session as it changes
• Use structured session information when coordinating work across agents or threads

That seems much more powerful than making every agent independently figure out how to inspect T3's SQLite database or provider logs.

Ideally, the underlying session API could power both the CLI and a T3-native agent tool/skill. That would give humans a useful command-line interface while also giving agents a supported, structured way to understand the T3 environment they're operating in.

The existing t3-session project could serve as a reference implementation for the behavior and data that has proven useful. I don't think T3 Code necessarily needs to depend on or directly incorporate that package, but having an officially supported equivalent in T3 itself would be extremely valuable.

Reference implementation: https://github.com/Albro3459/t3-session

[Albro3459]

Looks like #2829 would resolve this