docs: add guidance for intentional ask_user pauses

[PayalVerma1]

Summary

• Added author-facing guidance explaining how to intentionally pause execution for user input using ask_user.
• Clarified that interactive services are authored by leaving required inputs unbound, not by introducing a ### Runtime: interactive: true field.
• Added a complete interactive example demonstrating bind-time ask_user behavior and linked it from the examples index.

Fixes #105

---

Why

The existing documentation describes ask_user only as an automatic VM input-binding mechanism. It does not clearly explain how authors intentionally build interactive services, making this behavior difficult to discover.

As a result, contributors inferred a non-existent ### Runtime: interactive: true option because there was no author-facing guidance or interactive example to follow.

This PR closes that documentation gap without changing language semantics or runtime behavior.

---

What Changed

Documentation

• Added an Interactive Authoring section to skills/open-prose/guidance/authoring.md explaining:

  • how to intentionally pause for user input,
  • when ask_user should be used,
  • expected behavior for non-interactive hosts (unresolved-intent rather than fabricated input).

• Added Intentional ask_user Pauses to skills/open-prose/prose.md, documenting the recommended authoring pattern:

  1. Declare the required input.
  2. Leave it unbound.
  3. Allow the VM to invoke ask_user, bind the response, and resume execution.

  The section also explicitly clarifies that ### Runtime: interactive: true is not part of the language.

Examples

• Added a new interactive-grill example showing an intentionally interactive responsibility that pauses for user input.
• Added supporting documentation explaining bind-time pauses and non-interactive fallback behavior.
• Linked the new example from the examples index as the interactive counterpart to the existing non-interactive examples.

---

Design Boundary

This PR is documentation and examples only.

It does not modify:

• language semantics,
• VM behavior,
• runtime execution,
• parser/compiler logic,
• host adapters.

---

Testing

• ✅ pnpm build
• ✅ pnpm test:skill

Results:

• 27 test files passed
• 411 tests passed

---

Follow-ups

This PR intentionally addresses only the documentation/discoverability gap described in #105.

The host-specific interaction behavior (structured vs. plain-chat questioning) and the potential run-log improvements mentioned in the issue remain separate concerns and are better addressed in future issues or PRs.