rcan 3.6.0

Added

• MessageType.AUTHORIZE = 45. Spec §16.4 requires an AUTHORIZE message but the canonical table never numbered it; 45 is the next free value (RobotRegistryFoundation/rcan-spec#221). Additive: no existing value changes.

• rcan.assurance: helpers for RCAN Appendix C, the Physical Assurance Profile
  (informative, optional; rcan-spec PR #221, not yet merged). Python port of the
  rcan-spec reference verifier scripts/assurance/evidence-chain.ts, with the
  same hashing and finding codes:

  • GateDecision and Envelope TypedDicts, the Decision literal
    (allow | clamp | reject | stop), Finding, ASSURANCE_LEVELS
    (A1, A2, A3) and GENESIS_PREV.
  • envelope_hash, record_hash and append_record, built on
    rcan.encoding.canonical_json.
  • verify_chain(chain, expected_head=None) (EV-08: BAD_GENESIS, SEQ_GAP,
    PREV_MISMATCH, HASH_MISMATCH, HEAD_MISMATCH). Removal of records from
    the end of a chain is only detected against an anchored expected_head.
  • audit_authority (EV-07 log half: NO_PRINCIPAL, NO_AUTHORITY).
  • replay_against_envelope (ENVELOPE_MISMATCH, REJECT_APPLIED,
    STOP_WITH_MOTION, SPEED_EXCEEDED, TURN_EXCEEDED, OUTSIDE_KEEP_IN,
    INSIDE_KEEP_OUT, and UNCHECKED_FIELDS for fields it does not judge).
  • The functions and constants are also exported from rcan.

• Tests in tests/test_assurance.py with fixtures copied from rcan-spec. They
  recompute every fixture record hash and the envelope hash produced by the
  TypeScript reference, as a cross-language parity check.

These helpers verify evidence, not robots. Physical assurance levels A1-A3 are
independent of RCAN conformance levels L1-L4. Conformance is not certification.