v0.15.0

Per-kind incremental polish — the polish-cost-reduction lever 2.

Regeneration now skips template kinds whose deterministic pre-polish scaffold is unchanged vs. what produced the on-disk file: no polish LLM call, the polished body is kept verbatim, and deterministic metadata (source_hash/generated_at) is refreshed in place so staleness still clears. A feature where 2 of 11 kinds' content actually changed makes exactly 2 polish calls instead of 11.

• New scaffold_hash frontmatter field (help templates) / footer attr (project docs)
• --overwrite bypasses the skip; pre-0.15.0 files regenerate once then carry the hash
• Synchronous and batch paths inherit the skip identically

Spec: attune-ai docs/specs/polish-cost-reduction/ (D3). Full details: CHANGELOG 0.15.0