v0.26.10 - pk-doctor false-actionable cleanup

v0.26.10 is a patch release that clears false-actionable
pk-doctor findings after the v0.26.9 preauth fix and hardens ID token
reservation.

Fixed

• Downgraded historical lexical-token collisions to non-actionable
  info. Existing timestamp/word-pair shorthand collisions now report
  as historical context instead of open migration work, while new ID
  generation continues to avoid reusing reserved lexical tokens. Closes
  #51.
• Reserved lexical tokens across all ID generation paths.
  id-management.generate_id() and the shared processkit ID helper now
  reserve every lexical token already present in the target kind, so
  pair, intent, and generic ID generation all avoid ambiguous reuse.
• Stopped rejected migrations from keeping archive hygiene
  actionable. The pk-doctor applied-migration archive check now counts
  only state = applied migration records, so rejected records can stay
  available as explicit policy dispositions.

Verification

• pk-doctor: 0 errors, 0 warnings, 0 actionable infos
• pk-release-audit --tree=both
• uv run scripts/smoke-test-servers.py
• Focused pytest coverage for ID vocabulary and ID generation