purrdf 1.1.0

The first release after 1.0.0. Two reported bugs, the release fallout 1.0.0 left
behind, and a data-conflation bug in the GTS carrier found while coordinating
with the upstream wire-format freeze.

Additive: no published API changes shape. A TermRow widening that would have
forced a major bump was reworked into a parallel column before release — RDF 1.2
base direction is now purrdf_rdf::gts_view::term_directions() and a
directions key on the Python projection dict, so every 1.0.0 caller keeps
working.

Bug Fixes

• shacl: sh:message templates are substituted on the sh:sparql path.
  {$path}/{$value} shipped as literal braces because the constraint's
  message was resolved once and cloned into every result row; it is now rendered
  per solution, from that solution's own bindings. Both {?var} and {$var}
  spellings, and {$this} whether or not the SELECT projects it. An unbound
  placeholder is still left verbatim rather than blanked.
• cli: purrdf validate resolves a shapes graph's owl:imports from
  --import IRI=FILE, transitively and cycle-safely. Previously they were
  ignored in silence, so a shapes graph whose shapes lived in its imports
  reported conforms true against no shapes at all. PurRDF fetches nothing: an
  unresolved import with no --import given is reported on stderr, and with a
  table given is refused by name.
• gts: RDF 1.2 literal base direction is part of a term's identity in the
  snapshot composer. It was absent from the intern key, so "Cat"@en--ltr and
  "Cat"@en--rtl merged into ONE term and every quad was repointed at the
  survivor — a different graph, not a missing column. Emitted bytes are
  unchanged for any input without directional literals.
• release: The crates.io bootstrap script packages and publishes its PLAN
  rather than the whole ledger; a deferred crate no longer reaches cargo package. Its self-test asserts the arguments handed to the irreversible step,
  and the arm that had been failing since the ledger emptied is fixed — that arm
  runs before cargo test in make check, so it had been failing the gate for
  every contributor.

Features

• python: gts_to_sqlite, gts_to_duckdb and gts_to_parquet are
  implemented. They previously raised unconditionally. Five tables in the
  projection's own row order, so the same container exports to the same content
  twice. SQLite needs only the standard library; the other two take the
  [duckdb] / [parquet] extras.
• rdf: gts_view::term_directions() exposes RDF 1.2 base direction from the
  relational projection, which previously dropped it.
• entail: entails::imports::imported_iris() is public, so a consumer can
  read a document's owl:imports without re-deriving which objects count.

Documentation

• pack: The pack identity digest is no longer captioned "RDFC-1.0". It is
  computed through the purrdf-rdfc12 profile, which agrees with RDFC-1.0 byte
  for byte only on the RDF 1.1 subset. Labelling only — no digest bytes change.

Build

• ci: Benchmarks moved off the per-push path to a weekly schedule plus
  workflow_dispatch, and its uv installer is pinned.
• make: make doctor reports which build pins the local machine actually
  enforces, and the wasm gate distinguishes "rustup absent" from "target not
  installed" instead of printing the same line for both.
• python: bench = false on the binding lib, so cargo bench --workspace
  links.