v0.8.2

[0.8.2] - 2026-09-12

Fixed

• Creating state walks reprocess every opened directory link before descent,
  including existing directories left by failed attempts. New pairs sync child
  then parent unconditionally; existing pairs qualify each descriptor separately
  using fd-based readonly mount flags. All mount-query and writable sync errors
  fail closed before descendant creation, record publication or lock actions.
• mkdirat races returning EEXIST no longer trigger intermediate-directory
  chmod. Final private-directory repair and existing security/atomicity checks
  remain intact; non-creating walks retain their prior behavior.

Added

• Offline synthetic coverage for all-link ordering, repeated failed retries,
  readonly/writable descriptor combinations, mount-query/sync failures,
  deterministic creation races and write/lock failure barriers.

Documentation

• Disclosed extra synchronization and possible storage errors on writable
  ancestors previously unchecked. Conditional recovery assumes stable topology,
  storage honoring sync and independently durable readonly namespaces;
  ST_RDONLY is not persistence proof. No physical power-loss simulation,
  universal mutation-journal guarantee, journal or POST capability is claimed.
  Future mutation activation still requires persistent-storage eligibility,
  policy/outcome safeguards and separate live consent.