Include Pick & polish in skill-based feedback proposals #76

Prototype authors need directed human requests and corrections from Pick & polish included when they run /review-hub-feedback and hand accepted findings to /prepare-feedback-proposals. These conversations should use the same independent relevance and transfer standards as ordinary review comments, without extra weighting or automatically turning one correction into a rule.

Retrieve complete conversations through the existing read-only API, preserving namespaced source identities, original Build and target, author roles, follow-ups, corrections and contradictions. Agent answers provide context, not independent support or verified resolution. Compare storage and retrieval metadata with ordinary comments and fix only gaps needed for reliable skill consumption; document intentionally absent metadata.

Keep private evidence and approval-gated proposal application. This does not change automatic Hub Learnings, packages/learning ingestion, UI behavior, scheduling or publishing infrastructure. This work builds above #71 and the proposal skills merged in #67.