The #6408 request-shape-keyed TTL cache around getUnifiedModelsResponse was not
keyed by DB/settings state, so a write followed by a read within the ~1.5s TTL
replayed the pre-write catalog (dropping newly-eligible models like codex/gpt-5.5
and every specialty catalog routed through it since #6303). Fold a
modelCatalogCacheVersion (bumped by invalidateDbCache, already called on every
settings/connection/combo/pricing write) into the cache so a state change forces
an immediate miss; merge response headers through a real Headers instance
(fixes X-Request-Id duplication); carry and replay status end-to-end (a mid-build
500 was replayed as 200). Tests call the existing __resetCatalogBuilderRunsForTest
hook in setup, matching v1-models-concurrent-6408.
Allow image generation requests to omit prompts for models that only
accept image input, and validate required inputs from model metadata
instead of enforcing a text prompt for every request.
Treat authless search providers as executable with built-in defaults so
SearXNG can run without stored credentials, including during provider
auto-selection.
Also align runtime support with Node.js 24 LTS, harden thinking tag
compression and proxy wildcard matching, and update tests for the new
route and runtime behavior.
