docs: clarify state/preprocessor guidance for 0.7.8

[rlippmann]

What changed

• Clarified state opacity guidance in README.md:
  • engine.state is documented as an opaque authoritative snapshot.
  • normal host reads are directed to get_premise_value(...) and get_policy_items(...).
• Clarified preprocessor boundary guidance in README.md:
  • pending-clarification bypass (engine.has_pending_clarification() => pass raw input directly to engine.step(...)).
  • explicit false-negative-preferred abstain policy.
• Added the same pending-clarification bypass note to integration docs where preprocessor flow is described:
  • examples/integrations/litellm/README.md
  • examples/integrations/openwebui/README.md
  • examples/integrations/litellm_proxy/README.md
• Updated Demo 09 helper usage to public API style without behavior change (get_policy_items(...)).
• Bumped package version from 0.7.7 to 0.7.8 and updated uv.lock.

Why

• Keep release docs aligned with intended host usage boundaries (opaque state reads, conservative preprocessing flow, pending-clarification safety).
• Include release version metadata updates in the same PR for 0.7.8.

Checklist

• pre-commit run (uv run pre-commit run --all-files)
• tests pass (uv run pytest)