Restore dropped term-info fields; harden License + heavy-term paths

[Robbie1977]

What

Reconcile term_info_parse_object (vfb_queries.py) with the canonical
dataclass serialiser (term_info_queries.py) so the VFBquery get_term_info
path drops no fields the legacy SOLR term_info-field path renders, and fix two
operational issues found while running the parity harness.

Why

This is the Phase-2 gate for the term-info side-panel migration (legacy SOLR
term_info read → VFBquery get_term_info). Display parity is the gate. Three
reference-bearing fields were silently dropped:

• Class definition references (def_pubs) — never read. Every class page lost
  its definition citations (e.g. medulla FBbt_00003748 lost FBrf0231227,
  FBrf0224194). The legacy panel appends these inline to the definition
  (VFBProcessTermInfoCachedJson.java:937), so they are restored inline on
  Meta.Description, not as a separate Publications section.
• Individual synonyms (pub_syn) — the synonym block was gated
  "Class" in SuperTypes, so Individual terms (e.g. VFB_00101385) lost their
  synonyms. Class synonyms were already correct.
• Publication external content (pub_specific_content) — gated on SuperType
  "Publication", but the SOLR marker is the lowercase "pub"; the block never
  fired, dropping pub title / PubMed / DOI / FlyBase links (e.g. FBrf0242477).

Changes

• vfb_queries.py term_info_parse_object:
  • append def_pubs microref links inline to Meta.Description (matching the
    legacy definition render), instead of a structured Publications entry;
  • emit Synonyms from pub_syn for any SuperType, not Class-only (each
    synonym already carries its own publication inline);
  • gate the publication block on "pub"/"Publication" (pub title + links).
• vfb_queries.py fill_query_results: skip the limit=-1 full re-run used only
  to length-check a query when the preview was not saturated (preview rows < cap
  ⇒ count = preview rows). Cuts cold latency on SuperTypes offering many queries
  and the zero-count grey-out path. Saturated previews still re-run.
• solr_result_cache.py cache_result: default the cache write to a soft commit
  (commit=false) so a wedged IndexWriter cannot stall a cold-miss request and
  surface as ha_api 503 (the License-term symptom). Override with
  VFBQUERY_SOLR_WRITE_COMMIT=true.
• src/test/test_term_info_parity.py: parity tests for the three field gaps plus
  a License smoke test.

How to test

VFBQUERY_CACHE_ENABLED=false python -m pytest src/test/test_term_info_parity.py -v

All six pass. Parity completeness harness
(projects/term-info-vfbquery-migration/parity_harness.py) reports 0 dropped
identifiers across the full SuperType set.

Follow-ups (out of scope)

• Deeper count optimisation: dedicated count(...) / Owlery count modes per
  heavy query function (this PR only avoids the unnecessary re-run).
• Pre-warm License / non-query-reachable SuperTypes in owlery-cache-reload.

Release

Minor bump to v1.22.0 (tag-driven). After merge:
owlery-cache-reload --only "V3 term info" --force-refresh.

[Robbie1977]

Pushed a follow-up commit from a pre-release field-coverage sweep against the legacy processor (VFBProcessTermInfoCachedJson.java). Beyond the three named gaps, term_info_parse_object also dropped:

• xrefs — external DB links (medulla Insect Brain DB; gene FlyBase). Now a structured Xrefs list.
• related_individuals — present on most FBbt classes. Now Meta.RelatedIndividuals.
• targeting_splits / target_neurons — wired (TargetingSplits/TargetingNeurons) to match the legacy model; unpopulated in current data so a no-op today, but no longer droppable.

New fields declared on TermInfoOutputSchema. images/downloads/queries already covered (image records carry nrrd/obj/wlz/swc + template geometry). Tests now 9/9; harness spans 12 SuperTypes, 0 dropped.

[Robbie1977]

Reviewed VFB_json_schema_indexer (the indexer producing the SOLR term_info field) — its vfb_query_builder QueryLibrary *_term_info() methods are the authoritative per-SuperType clause set. Cross-checked every clause against term_info_parse_object:

Now covered (this PR): term/parents/relationships/synonyms(pub_syn)/def_pubs/xrefs/related_individuals(+term_replaced_by)/license/pubs/pub_specific_content/channel+anatomy images/template channel+domains, and newly DataSet term.link/term.logo → Meta.Link/Meta.Logo.

Query-derived, not static fields — removed from this PR, tracked as a separate follow-up:

• targeting_splits (Neuron classes) / target_neurons (Split classes) — Cypher in the indexer neuron_split/split_neuron clauses; these become query types with a count badge.
• dataset_counts (images/types) — already served by the DatasetImages query count.

Tests 10/10; harness 12 SuperTypes, 0 dropped.

[Robbie1977]

Added targeting_splits/target_neurons as live query types (your call — they're query data, not static fields):

• get_splits_targeting (Class+Neuron) and get_neurons_targeted_by_split (Class+Split) — live Neo4j queries mirroring the indexer neuron_split/split_neuron clauses, returning the standard class-row table + true count, wired into term_info_parse_object and QUERY_TYPE_MAP with a count badge.
• Verified live: MBON FBbt_00100243 → 33 targeting splits; split VFBexp_FBtp0129935FBtp0129968 → 18 target neurons.

Also in this PR: DataSet term.link/term.logo → Meta.Link/Meta.Logo (License url/logo too). Tests 14/14; harness 12 SuperTypes, 0 dropped. Holding for your review before merge.