Refactor schema workflows to avoid extra operations and duplicate sch…

[satyakigh]

Problem

Duplicate schema fetches and tight coupling between SchemaRetriever, GetSchemaTaskManager, and SchemaStore via callbacks and direct datastore access.

Solution: Architectural Simplification

SchemaStore (New API Layer)
• Added controlled getters: getPublicSchemas(), getSamSchemas(), getPrivateSchemas(), getPublicSchemaRegions()
• Added age tracking: getSamSchemaAge(), getPublicSchemasMaxAge()
• Smart caching in get(): only rebuilds combined schemas when region changes
• invalidate() method clears cache when underlying data updates
• Eliminates direct datastore access from other components

GetSchemaTaskManager (Simplified)
• Removed: Settings subscription, automatic background timers (10s/1hr), onSchemaUpdate callback, profile tracking, Closeable interface
• Changed: Now directly calls schemas.invalidate() after private/SAM tasks complete
• Added: processedRegions Set for deduplication (tracks regions ever processed)

SchemaRetriever (Delegated)
• Added explicit initialize() method that checks schema staleness (7-day threshold)
• Delegates to GetSchemaTaskManager instead of managing callbacks
• Preserves firstCreatedMs when refreshing stale schemas

Key Behaviors

Startup: Constructor immediately calls addTask(defaultRegion), then initialize() checks all cached schemas for staleness and triggers refreshes

Caching: Combined schemas cached by region+profile key, only rebuilt on region change or explicit invalidation

Fallback: Requesting unavailable region returns last available region's schemas (graceful degradation)

Deduplication: processedRegions Set prevents re-adding same region, but never clears entries

Test Changes

• New comprehensive SchemaStore.test.ts
• Modernized tests using sinon stubs instead of vi mocks
• Better async handling with waitFor() helper