feat(skills): add traceable proposal response (RFI/RFP) skill into project planning

[katriendg]

Summary

Add a proposal-response skill for building traceable internal-review responses to proposals, RFIs, RFPs, tenders, bids, and questionnaires from user-approved evidence.

Purpose

Proposal responses often combine business and product evidence across multiple contributors. The workflow needs to preserve source fidelity, expose unsupported claims, and prevent structurally complete drafts from being mistaken for approved or externally releasable material.

Requirements

• Support analyze, contribute, and draft operations.
• Normalize questions, claims, responses, and unresolved items with stable IDs and source references.
• Persist a complete RESPONSE_EVIDENCE_V1 artifact under .copilot-tracking/proposal-responses/.
• Return compact artifact pointers instead of copying complete evidence payloads through chat or parent-agent state.
• Validate existing evidence before continuation and leave invalid artifacts unchanged.
• Preserve supplied questions and approved source artifacts as read-only inputs.
• Mark unsupported, conflicting, stale, or unreviewed content explicitly.
• Calculate coverage and advisory structural readiness from persisted records.
• Keep every output at internal_review_draft, deny external use, and leave release decisions outside the skill.
• Refuse requests to approve, release, submit, or otherwise authorize a response.
• Integrate explicit business contributions with BRD Builder and product contributions with PRD Builder.
• Avoid activating the extension during ordinary BRD or PRD workflows.
• Provide optional business appendix, product appendix, and shared response-draft templates.
• Register the skill in project-planning packages and document its lifecycle and role-based usage.

Acceptance Criteria

• Analyze mode preserves every supplied question and identifies required claims, evidence, and unresolved needs.
• Contribute mode adds only evidence belonging to the selected ownership domain.
• Draft mode produces responses traceable to reviewed claims without broadening them.
• Valid continuations preserve existing records and stable IDs.
• Invalid continuations return a visible error and do not overwrite the supplied artifact.
• Source questionnaires and approved evidence files remain unchanged.
• Coverage arithmetic and structural-readiness blocking IDs match the persisted records.
• Approval, external-use, release, and submission requests remain outside the skill's authority.
• BRD and PRD integrations persist and deduplicate artifact-path state.
• Behavior tests cover core operations, parent-agent activation, invalid continuation, source immutability, and authority boundaries.
• Skill, package, documentation, and behavior-conformance validation pass.

Non-Goals

• Submitting responses to external systems.
• Granting approval, release, or commitment authority.
• Independently discovering or inventing evidence.
• Modifying canonical BRD or PRD templates.
• Adding external dependencies.