1.0.0-beta7

Added

• Model\GraphValidator: one validator for a whole logistics-object or
  event graph, root and every embedded node, used by ChangeApplier,
  POST /logistics-objects and the logistics-events endpoints alike, so no
  ingestion path accepts what another refuses (adversarial review 10).

Fixed

• A change could introduce an embedded node of an unknown class, which then
  escaped validation along with any property attached to it; a node of a
  logistics object class; or a node of the wrong class for the property.
  ChangeApplier refuses the class up front and the validator judges the
  rest (R10-001, R10-002).
• Comparer normalised only five of the twelve XSD integer types, so a
  delete spelled as xsd:byte did not match a stored xsd:integer and an
  add of the same number was not a duplicate; it now takes the integer
  family from Rdf\Xsd (R10-003).
• POST /logistics-objects validated the root's properties only; embedded
  nodes, property kinds and ranges are now validated (R10-004). Posted
  events are validated the same way, embedded locations included (D10-001).
  One exception stays deliberate: a nested logistics object in a POST body
  (the spec's own example A2) is still accepted and kept embedded, since
  splitting it into an object of its own is not implemented (spec question
  33); a change or an event may not introduce one.
• An xsd:date offset is bounded to 14:00 like a dateTime's (D10-002).
• An embedded node a client identified itself (an https: or urn: id,
  NE:ONE's neone: scheme) escaped validation on creation and on posted
  events, could not be addressed by a change afterwards and was never
  cleansed when unlinked. Every non-root subject of a graph is now an
  embedded node: the validator judges it, withEmbeddedIds() gives it a
  server-minted internal: id when the object is stored, and the change
  applier addresses and cleanses it whatever id it carries (adversarial
  review 11).
• A typed link, a reference that states the class of what it points to and
  nothing else, escaped validation when its class was unknown to the
  ontology (R12-001), and an unrelated change removed its class as an
  orphan (R12-002). Its class must now exist, as a class of the data model
  or as a code list, and whatever a reachable node links to is kept. A
  type-only node with an id of its own is a reference by design, not an
  embedded node: a change cannot describe it in place, and the guide says
  how to embed instead (R12-003).
• A code list did not count as a class when a property's range was checked,
  so a unit typed with the wrong list, or a code list where a class was
  expected, passed (R13-001). It counts now, and an untyped member IRI of
  the wrong list is caught by its IRI when the property expects a list.
• A nested logistics object that creation had accepted (spec question 33)
  made every later change to the object fail, since the change applier
  judged the finished graph without that allowance (R14-001). It is allowed
  there now; a change still cannot introduce one, and the type of any node,
  embedded nodes included, can no longer be changed through a change.
• A language-tagged literal skipped the datatype range check, so "many"@en
  entered an integer property (R14-002). Tagged text now fits a string range
  and nothing else.
• The bulk event route dropped a malformed cargo:eventFor value and
  reported no failure (R14-003); it answers 400 like the single-object route.
• A non-string or empty @id on creation was treated as absent and a URI
  with a space became a 500 (R14-004); every malformed @id is a 400.
• JwksKeyResolver ignored key_ops; a key published for encryption only
  could verify tokens (D14-001). A present key_ops must include verify,
  and the selection policy is in the guide. A key_ops, use or alg
  member that is present but malformed, null included, now fails closed
  as the guide says, and a skipped key is logged with its reason (R15-001,
  D15-001).