[Deferred] Generate VoiceClaw integration intent in NemoClaw main configuration

## Deferred from R0

R0 consumes hand-authored configuration directly in v1. #11751 owns that fixture and the apply-time onboarding trigger. This issue stays open for later main onboarding/configuration-export work; it is not an R0 child or prerequisite.

## Outcome

Main produces NemoClaw configuration declaring that one sandbox agent may accept voice. v1 applies the configuration and prepares the endpoint/credential handoff to externally managed VoiceClaw.

## Scope and ownership

The integration entry contains NemoClaw-owned intent: integration name, kind and one agent reference, scoped by the containing deployment UID. It does not contain VoiceClaw settings, credentials, endpoints, executable commands or installation hooks.

VoiceClaw owns its distribution, onboarding, configuration format and external processes. Release compatibility and qualification belong to the shared contract/support record; NemoClaw does not become a VoiceClaw package manager.

Planning reference: `VOICECLAW_NEMOCLAW_API_CONTRACT.md`, proposed `v1alpha1` agreement, §§1–3 and 7–9, with Appendices B–C. The requirements below apply the contract's ownership boundary to this milestone. R0 is a limited experimental profile; it does not claim the complete first-release agreement or accept #11204 wholesale. The user-selected temporary R0 profile uses a private endpoint and one scoped credential instead of the proposed offer/lease exchange.

## Work

- Extend main's configuration model and generation/export path with the minimal voice integration declaration from contract §3.
- Reuse NemoClaw's agent, inference and policy configuration. Produce the document without provisioning or modifying an agent or runtime.
- Account for current main behavior: normal onboarding has runtime effects and existing config export reads an existing sandbox. Provide a separate path that authors the required intent without executing deployment steps.
- Validate the declared target and compatible schema. v1 revalidates actual target identity, ownership and support when applying and admitting.
- Publish fixtures for the v1 consumer in #11751, with explicit translation if its input schema differs.
- Document that the configuration requests voice access after the agent is ready; the temporary R0 runtime handoff is a private endpoint and scoped credential; it does not request NemoClaw ownership of VoiceClaw's lifecycle.

## Acceptance and validation

- [ ] Configuration can be produced with VoiceClaw absent and without deployment effects.
- [ ] One integration identifies exactly one agent within its sandbox/deployment.
- [ ] VoiceClaw's internal settings, credential schema, commands and endpoints are absent from the integration entry.
- [ ] Invalid target references and unsupported schema are rejected with actionable errors.
- [ ] The v1 consumer interprets the fixture consistently, and actual ownership is established only during apply.
- [ ] Ordinary omission of the integration is non-destructive; explicit removal behavior is documented.

## Dependencies

R0 parent epic: [#11746](https://github.com/NVIDIA/NemoClaw/issues/11746).

VoiceClaw counterpart: [VoiceClaw #7](https://gitlab-master.nvidia.com/jarvis/voice-claw/-/issues/7).

Blocked by:

- [#11747: [R0] Specify the v1-to-VoiceClaw onboarding handoff and scoped connection fixtures](https://github.com/NVIDIA/NemoClaw/issues/11747)

## Category

Feature

Start with VOICECLAW_NEMOCLAW_API_CONTRACT.md, §§1–3 and 7–9 with Appendices B–C, then trace main’s existing configuration model and config export path. Review #11751’s consumer fixture and the blocked #11747 handoff requirements. Done means the deferred export path produces and validates the minimal one-agent intent without deployment effects, while the listed acceptance checks pass.

Written by the indexing model from the issue text.