Perf: cache the ontology list, removing ~97% of /search response time

[mdorf]

Summary

Production tracer data from #243 (deployed in v7.4.3) shows restricted_ontologies averages 876 ms of the 899 ms /search pipeline (~97%), and the search_scoped attribute revealed why both request styles pay it: the scoped path (98% of traffic) loads all ~1,200 ontologies with full attributes via ontology_objects_from_params and selects the requested ones in Ruby, while the site-wide path loads them all directly. Full baseline: #244 (comment).

This PR introduces OntologyListCache, a process-level TTL cache of the loaded ontology list, and points the shared helpers at it. Expected effect: /search average drops from ~900 ms to roughly the 15-60 ms the rest of the pipeline costs; annotator, recommender, mappings, analytics, and properties search shed the same per-request tax.

Addresses #244.

Design (agreed with @mdorf)

• In-process, per-worker cache of the loaded Ontology objects; no Redis (goo objects don't marshal cleanly; rebuilding objects re-pays the materialization this removes).
• TTL configurable: LinkedData::OntologiesAPI.settings.ontology_list_cache_ttl, default 300 s; 0 disables caching entirely (debug kill switch). Override per environment in the deployment config if needed.
• Staleness contract: ontology add/remove and ACL/viewingRestriction changes take up to TTL per worker to appear. Per-user visibility (slice, custom sets, ACL via filter_access) is still computed per request; only the raw list is cached.
• Defensive copies: every read returns a dup because get_term_search_query filters the array destructively. Cached objects are shared within a worker: bring on them is safe, attribute assignment is not (documented in the module).
• Miss/expiry behavior = current behavior: same load call as today, executed once per TTL window instead of per request. Load failures propagate and are never cached.
• The load includes the attribute union all consumers need (serialization defaults + access-control attrs + viewOf), which lets the scoped branch drop its second query and moves the site-wide view filter into Ruby.

Observability

OntologyListCache.all (per-request, ~0 ms on hit) and .load_ontologies (fires only on miss) are added to the #243 tracer registry, so cache behavior is directly visible in production segments; the registry regression test covers them automatically.

Evidence

• Production baseline (18 h): restricted_ontologies 876 ms avg; scoped 962 ms / site-wide 878 ms transaction averages.
• Local microbenchmark (test backend, small dataset, directional only): cold load ≈ old per-request cost; warm hit 0.008 ms.

Test plan

• New test/lib/test_ontology_list_cache.rb (8 tests): caching across calls, defensive-copy isolation against destructive callers, TTL expiry (deterministic clock swap), TTL=0 bypass, default value, explicit invalidation, failed-load-not-cached, and a real-loader test asserting all consumer-needed attributes are preloaded.
• Test harness invalidates the cache per test method (tests create/delete ontologies at will; test/test_case.rb).
• Green locally: search (13 tests/147 assertions), search models, properties search, annotator, recommender, mappings, ontologies, slices, application helper, tracer registry.

Rollback / kill switches

Set ontology_list_cache_ttl = 0 (config + restart) to disable the cache without reverting; NEWRELIC_CUSTOM_TRACERS=off remains available for the tracers themselves.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 78.76%. Comparing base (4b26187) to head (5753d3c).

Additional details and impacted files

@@             Coverage Diff             @@
##           develop     #245      +/-   ##
===========================================
+ Coverage    78.60%   78.76%   +0.15%     
===========================================
  Files           67       68       +1     
  Lines         3754     3777      +23     
===========================================
+ Hits          2951     2975      +24     
+ Misses         803      802       -1     

Flag	Coverage Δ
ag	78.76% <100.00%> (+0.15%)	⬆️
fs	78.76% <100.00%> (+0.15%)	⬆️
gd	78.76% <100.00%> (+0.15%)	⬆️
unittests	78.76% <100.00%> (+0.15%)	⬆️
vo	78.76% <100.00%> (+0.15%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.