flow-next v3.8.0

Changed

• /flow-next:interview now tags the acceptance criteria it writes. Capture
  has always stamped provenance on every criterion ([user] / [paraphrase] /
  [inferred] / [strategy:<track>]), but interview emitted nothing - so the
  grounded-vs-guessed tally and the targeted "re-interview only the [inferred]
  items" pattern silently returned empty on any interview-authored spec, which is
  most of them for teams using the symmetric PO -> tech-lead flow. Interview now
  emits the same four tags at all three write-back sites, with [user] meaning
  the PO under a business pass and the tech lead under a technical one. A pass
  tags only the criteria it authors and never retags an existing bullet, so
  provenance is frozen exactly like the R-ID number; untagged criteria mean
  unknown provenance, never [user]. Capture's no-self-blessing rule is
  inherited, narrowed to [inferred] criteria that no question covered. Tag
  definitions are repeated at the emission sites rather than centralised
  (relocating them regressed accuracy once before) and a test pins the vocabulary
  as set-equality across both skills, so a renamed, dropped, or added tag fails
  CI in either direction. No parser or CLI change: the tag format was already
  source-agnostic.

Fixed

• The published source-tag tally recipe dropped every [strategy:*]
  criterion. The recipe shipped in 3.7.0 used the character class [a-z:]+,
  but a track name keeps its literal casing and may contain spaces or hyphens
  ([strategy:Cross-platform parity]), so those criteria never matched. The
  awk stage had a second, quieter version of the same bug: with the default
  whitespace split, a spaced track name landed in the wrong field and produced a
  phantom tag. Corrected in the repo docs and on flow-next.dev to [^]]+ with a
  tab-delimited awk -F'\t', and the reason is documented inline so it does not
  regress.