Upgrade horned-owl 1.4 -> 3.x and drop the [patch.crates-io] fork (unblocks publishing owl-dl-cli / cargo install rustdl)

[micheldumontier]

Why this exists

crates/owl-dl-cli is publish = false and crates/rustdl ships no [[bin]], and the recorded
reason was that the Manchester-syntax reader was "not yet in upstream horned-owl". That is no
longer true and had not been true for weeks. phillord/horned-owl#176 merged on 2026-07-11, and
io::omn ships in released horned-owl 2.0.0 (2026-07-17) and 3.0.0 (2026-08-10). The stale
claim is corrected in-tree as of e4a857d.

What actually blocks the CLI is narrower and more concrete:

1. this workspace pins horned-owl = "1.4" and io::omn first appears in 2.0, so dropping the
   [patch.crates-io] fork is a major-version upgrade, 1.4 -> 3.x; and
2. one item the fork carries did not go upstream.

What is present upstream vs. fork-only

Probed against docs.rs for 2.1.0 and 3.0.0:

path rustdl needs	upstream 3.0.0
io::omn::reader::read	present
io::omn::reader::parse_class_expression	present
io::omn::AsManchester	404 - fork-only

Upstream merged the reader plus a whole-ontology io::omn::writer::write. The fork's per-item
AsManchester trait did not land, and writer::write cannot substitute: it serialises an entire
ontology, whereas explanations render one axiom at a time.

AsManchester is reached through .as_manchester_with_prefixes(&pm) at 6 call sites in 3 crates:

• crates/owl-dl-cli/src/main.rs:1744, 1820, 1919, 1949 (justify / explain output)
• crates/owl-dl-cli/src/report.rs:85 (HTML report)
• crates/owl-dl-py/src/explain.rs:14 (Python explain surface)

Licence constraint: horned-owl is LGPL-3.0 and rustdl is Apache-2.0/MIT, so copying the
fork's implementation into this workspace is not clean. The options are to reimplement per-item
Manchester rendering over our own IR, or to upstream AsManchester to horned-owl.

Migration scale (measured 2026-09-03)

• 170 files, 976 horned_owl:: references, 49 distinct paths.
• Concentrated: the top four paths are 739 / 976 = 76% -
  ontology::set::SetOntology (222), io::ofn::reader::read (182), model::RcStr (171),
  io::ParserConfiguration (164).
• All six core paths rustdl depends on (SetOntology, ofn::reader::read, RcStr,
  ParserConfiguration, ForIRI, Build) were probed present at 3.0.0 - so the count is
  breadth, not necessarily breakage. The real work is wherever 2.0/3.0 changed signatures or
  semantics, which has not been surveyed yet. Do that first; it sizes everything else.

Scope of work

1. Diff upstream 1.4 -> 2.0 -> 3.0 for the 49 paths; produce a breaking-change list. Gate: do
   this before committing to the rest.
2. Decide AsManchester: reimplement over rustdl's IR, or upstream it.
3. Migrate, drop [patch.crates-io], drop the fork pin.
4. Flip owl-dl-cli to publish = true and add [[bin]] to crates/rustdl so
   cargo install rustdl works.

Evidence bar

This changes the parser under every engine, so it does not ship on a green cargo test. It
needs the full treatment this repo requires of any change to conversion:

• ./scripts/run-soundness-diff.sh - FP=0, all 11 curated closures exact.
• A 1,920-ontology two-arm ORE sweep with --digest-strip-comments, arm order alternated (a
  fixed order buys a ~3.4% page-cache phantom), single-thread, on an idle host - reporting answer
  identity, not just wall.
• Note the corpus is not self-validating here: a parser change can alter which axioms enter the
  KB at all. Adjudicate any DIFFER against Konclude/HermiT rather than assuming noise.