rustdl 0.3.0

[0.3.0] — 2026-06-05

Changed

• Classification now bounds each subsumption test by default
  (pair_timeout_ms / per_pair_timeout_ms defaults to 1000 ms;
  previously unbounded). Pathological SROIQ ontologies (e.g. pizza) no
  longer hang by default. A timed-out pair is recorded as "not
  subsumed" — sound (never a false subsumption), but the result may
  be incomplete. Pass 0 for the complete, unbounded classification.
  1000 ms is the empirical knee on pizza: higher budgets buy no extra
  completeness (the remaining pairs are intractable at any reasonable
  bound) but cost proportionally more wall time.
• Incompleteness is now surfaced loudly, so a bounded run can't
  silently hand back a hierarchy missing real edges:
  • CLI rustdl classify prints a prominent ⚠ INCOMPLETE: N pair(s) exceeded the timeout … warning to stderr when any pair times out.
  • Python rustdl.classify / classify_bytes emit an
    IncompleteClassificationWarning (filterable via the warnings
    module), and the Classification object exposes .complete (bool)
    and .timed_out_pairs (int).

Added

• CLI multi-format input. rustdl <cmd> file.{owx,owl,rdf} now
  works — the reader is chosen by file extension (.owx → OWL/XML,
  .owl/.rdf → RDF/XML, .ofn/other → OWL Functional). Previously the CLI
  read OFN only; the Python bindings already auto-detected. Verified on
  the owlcs pizza ontology (RDF/XML) — 99 classes, 2 unsatisfiable
  (CheeseyVegetableTopping, IceCream), matching the canonical result.