epic: agent-supported install path for NemoClaw

## Problem Statement

NemoClaw has a command-first quickstart and repo-shipped user skills, but the docs do not yet provide an up-front "copy prompt for your agent" install path. Users who prefer Cursor, Claude Code, Codex, Copilot, or another agent must discover the Agent Skills page, load the right skill, infer the setup flow, and recover from environment failures manually.

This is an agent-readiness gap for the priority workflow "install NemoClaw and launch a first sandbox." The workflow should let an agent discover canonical guidance, ask the right setup questions, execute with explicit user consent, verify readiness, and self-correct from known install/onboarding failures.

## Proposed Design

Create a Databricks DevHub-style agent-supported install path for NemoClaw:

- Add a first-class docs CTA, "Copy prompt for your agent," near the top of the home and quickstart experience.
- Ship the copied prompt from a repo-owned source artifact, not one-off page text.
- Make the prompt self-contained enough for users who have not cloned the repo or installed NemoClaw skills yet.
- Have the prompt direct agents to use NemoClaw existing user skills when available, especially `nemoclaw-user-get-started`, and to fetch/install/update the skills when missing or stale.
- Require the agent to verify prerequisites, explain tradeoffs, ask before destructive or privileged steps, run install/onboard with user consent, and validate success with a sandbox status check plus first-agent-prompt path.
- Align the flow with Agent Readiness Standards: workflow-first, discoverable, executable, self-correcting, packaged in the product repo, and backed by before/after validation.

## Sub-Issues

- [x] #5048
- [x] #5050
- [ ] #5051

## Acceptance Criteria

- A user can copy one prompt from the docs and paste it into a supported coding agent to begin NemoClaw installation.
- The prompt tells the agent to read first, ask one material question at a time, use structured multiple choice when available, and avoid unattended install behavior unless explicitly authorized.
- The agent flow covers prerequisites, installer consent, provider/model choice, sandbox name, optional integrations, policy tier, success verification, and troubleshooting.
- Existing generated skills are reused where possible; any new artifact has a clear owner, source file, generation path, and update story.
- Docs build passes and the copied prompt remains in sync with the repo-owned source.
- Validation evidence shows improved agent task success or reduced failure/recovery steps compared with the current docs-only path.

## Alternatives Considered

- **Keep command-first docs only:** simpler, but misses users who start from an agent and does not meet the agent-readiness intent.
- **Only link to Agent Skills:** useful but too indirect; users still need to know what to load and how to start.
- **Create a brand-new install skill only:** may duplicate the existing `nemoclaw-user-get-started` skill unless the new artifact is explicitly a compact bootstrap prompt.

## Checklist

- [x] I searched existing issues and this is not a duplicate.
- [x] This is a design proposal, not a "please build this" request.

## Notes

- GitHub CLI did not expose a native Issue Type setter in this environment, so this uses the `epic:` title prefix and routing labels.
- Treat the Databricks example as interaction inspiration, not as text to copy.