Importing an ontology queues alignment behind the index refresh, as the editor does

[Maya-Kid]

Importing an ontology now queues alignment behind the index refresh, so kind words and phrases decided earlier can bind to what it adds.

Problem

apply_import runs owl_import::apply, which queues only embed_ontology. The editor's create and update handlers (#956) and adoption (#1257) queue the refresh, then the alignment, as docs/design/ontology.md asks of every path adding elements. The new-base form says more can be imported later.

On dev, a document gives Acme the kind word company, decided none with no classes yet. After importing ex:Company a owl:Class and running embed_ontology, company stays none and Acme has no class until another document or a class edit queues alignment. Phrase signatures with no fitting property stay none too.

Change

• After the import's reconcile commit, apply_import calls reindex_then_align(&state, kb_id, "align_types"), which finds the refresh already queued and adds align_types; that job waits for the refresh and queues align_phrases when it settles. Not in owl_import::apply, which also installs packs into new bases that have nothing to align.
• The design doc lists imports too.

Verification

• api::token_setup_tests::importing_an_ontology_queues_alignment_behind_the_index_refresh imports a one-class Turtle file through the router and expects ["embed_ontology", "align_types"] queued. On dev 040a533 with only the test file added, it fails with ["embed_ontology"]. Cleanup now deletes the base's jobs.
• On dev 2293ec9, a development test ran the queued jobs: company bound to the imported class, and stayed none without the change. It is left out: past the queue it tests the unchanged aligner.

Linux x86_64 container, Rust 1.98.1, PostgreSQL 16 + pgvector, Poppler, DB required, commit 96d37a5 on dev 040a533: fmt and clippy clean; cargo test --locked --workspace 1,576 passed, 0 failed, 6 ignored.

Limits

• A failed import queues no alignment, though apply may have committed some classes; a successful re-import queues it.
• Re-importing an unchanged file re-asks the agent's kind-word decisions whose candidates include its classes, since update_type_from_import always bumps updated_at, as an editor save does; before, the next document did.
• A properties-only file imported into a classless base queues no phrase alignment, like a document there: align_types queues it only once a class exists.

[WaylandYang]

Thank you. Calling reindex_then_align after the reconcile commit gives import the same order the editor and adoption already have: the index refresh that apply queued is reused, and align_types waits for it and queues align_phrases when it settles. Leaving it out of owl_import::apply keeps pack installs into new bases from queuing alignment with nothing to align. The test goes through the real import route and fails on dev. Merging.