v0.1.16

Added

• attune-rag-benchmark --compare-thinking runs the judge
  twice (thinking off, thinking on at --thinking-budget) and
  prints a side-by-side aggregate table plus a per-query
  verdict-shift list. Mutually exclusive with --thinking and
  --native-citations. Calibration data lives in
  docs/rag/faithfulness-thinking-calibration.md (resolves
  #17).
  Outcome of the 2026-05-15 calibration run: 80 % verdict-shift
  rate but mean faithfulness barely moves (−0.005) and
  hallucination rate worsens slightly (+2.5 pp); --thinking
  stays opt-in pending hand-labeled ground-truth queries.
• attune-rag-benchmark --json PATH dumps the full
  structured faithfulness report — including per-query
  reasoning text and the supported / unsupported claim lists —
  to a JSON file for offline analysis. Per-query benchmark
  records also gain supported_claims, unsupported_claims,
  reasoning, and latency_ms fields.
• plan_rename(..., kind="template_path") is now
  implemented (was NotImplementedError in v0.1.15). Moves a
  template file within its corpus root and updates path-keyed
  sidecars (summaries.json / summaries_by_path.json) when
  present. RenamePlan grows a new moves: list[FileMove]
  field and FileMove(old_path, new_path) is exported from
  attune_rag.editor. apply_rename applies moves first
  (creating missing parent directories, tracked for rollback)
  then text edits, reversing prior work on any mid-flight
  failure. Out of scope: cross-corpus moves, cross_links.json
  updates, attune-help static index, git history. The gui-side
  "Rename file…" trigger and WS path-change handling live in
  attune-gui. See docs/specs/template-path-rename/.