v0.3.4

Fixed

• TriggerRebuild no longer passes $fingerprint to FinalizeIndex. The $fingerprint parameter was removed from FinalizeIndex's constructor in the 0.3.0 rewrite, but TriggerRebuild was not updated. The md5 hash was silently flowing into the $memoryBudget slot and causing auto-rebuild jobs to always use the conservative memory budget, ignoring any explicitly configured budget.
• Hygiene: Replaced md5(serialize($items)) with md5(json_encode($items, JSON_THROW_ON_ERROR)) in TriggerRebuild for content fingerprinting — json_encode is faster, produces deterministic output across PHP versions, and avoids serialize baggage.
• Hygiene: Added === false error check to file_put_contents in DownloadPagefindCommand — failed .env writes now report an error instead of silently continuing.
• Hygiene: Added TOCTOU-safe comments to intentional @unlink calls in CleanupCommand and DownloadPagefindCommand.
• Hygiene: Added source-parse tests preventing reintroduction of bare serialize() and unchecked file_put_contents.

Added

• LaravelCacheDriver behavior tests. New ScoltaCacheBehaviorTest: verifies the driver contract (get/set/miss/array values) and end-to-end handler+driver caching — second call to handleExpandQuery/handleSummarize serves from the in-memory Cache facade (AI called once), while cacheTtl=0 calls the AI service both times. Uses Cache::swap(new Repository(new ArrayStore)) so no real cache store is needed.
• New ScoltaConfigIntegrationTest. Verifies the full pipeline: Laravel config → ScoltaAiService::flattenConfig() → ScoltaConfig::fromArray() → toJsScoringConfig() / toAiClientConfig(). Covers all 8 scoring fields, language, recency_strategy, all 5 display fields, feature toggles, AI client config (provider/model/base_url/omission), ai_languages, custom_stop_words, phrase proximity (defaults + overrides), cache_ttl (including 0), and all three prompt overrides.
• AI configuration tests (Phase 5). Added test_ai_feature_toggle_defaults asserting that ai_expand_query, ai_summarize, and max_follow_ups have the correct default values (true, true, 3).
• Scoring behavior tests (Phase 1). Extended test_scoring_section to assert language, recency_strategy, and recency_curve keys are present in the scoring config. Extended test_scoring_defaults to assert their default values. Added test_ai_languages_default confirming ai_languages is an array containing 'en'.