Add terminating tool results and defineTool helper

[HabibPro1999]

What

• Add terminate?: boolean support to tool results/messages.
• Suppress only the automatic model continuation when every finalized tool result in a batch terminates.
• Preserve queued Agent.followUp() handling after terminating tool batches.
• Add the typed extension SDK defineTool() identity helper.
• Update changelogs for the user-facing API additions.

Why

This ports two small Pi runtime-harness improvements to OMP:

• deterministic tools such as approval/yield/final-answer style tools can end the current turn without an unnecessary follow-up LLM call
• extension authors get a typed helper for custom tool definitions and Pi-style examples

Part of the Pi runtime/agentic harness migration stack.

Testing

• bun test packages/agent/test/agent.test.ts packages/coding-agent/test/extensions-runner.test.ts
• bunx biome check on all files changed by this PR
• Earlier pre-rebase checks: bun run check:ts and PATH="$HOME/.cargo/bin:$PATH" bun run check

Known caveat after rebasing onto latest main: full bun run check:ts is currently blocked by unrelated Biome formatting/import-order issues already present in origin/main files not touched by this PR (line-hash.ts, tools/renderers.ts, tools.test.ts, tools/search-renderer.test.ts).

---

• bun check passes (blocked after latest main rebase by unrelated upstream Biome issues listed above)
• Tested locally
• CHANGELOG updated (if user-facing)

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 887b4f4f58

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Keep follow-up queue processing after terminate batches

Breaking out of the outer loop here skips the getFollowUpMessages() check entirely when all tool results in a tool-use batch have terminate: true. That regresses the Agent.followUp() contract (“processed after the agent finishes”): follow-up messages queued while streaming can remain stuck in the queue with no automatic delivery, because the run ends before the follow-up dequeue path is reached. This shows up when a terminating deterministic tool finishes and the user (or session code) has already queued follow-up input during that same run.

Useful? React with 👍 / 👎.

[HabibPro1999]

Addressed the Codex follow-up review in 4c0f8ab.

Terminating tool batches now suppress only the automatic model continuation; they still fall through to the normal follow-up dequeue path so queued Agent.followUp() messages are delivered.

Added regression coverage for a follow-up queued during a terminating tool execution.

Verification:

• bun test packages/agent/test/agent.test.ts — 15 pass
• bun run check:ts — pass