feat(ai): on-device RAG over bookmarks + Spotlight indexing + unified logging

- SpotlightIndexer: actually push BookmarkEntity into the Spotlight index via
  indexAppEntities (IndexedEntity conformance alone indexes nothing). Wired into
  every sync/mutation point. This is what lets Apple Intelligence answer
  free-form Siri questions grounded in the user's bookmarks.
- On-device RAG: BookmarkAssistant (LanguageModelSession) + BookmarkSearchTool
  + SpotlightBookmarkSearch (CSSearchQuery retrieval), surfaced via AskView with
  lightweight Markdown rendering of answers.
- Log: os.Logger facility (com.magicive.marks) replacing ad-hoc print(); shared
  with ShareExtension. ClaudeService now logs server error bodies on non-2xx.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Marks/Intents/MarksAppIntents.swift

@@ -133,6 +133,13 @@ struct SummarizeBookmarkIntent: AppIntent {
     }
 }
 
+// Note: there is intentionally no "Ask" AppIntent. Apple Intelligence already
+// answers free-form questions over the user's bookmarks by querying the
+// Spotlight index of `BookmarkEntity` (an `IndexedEntity`) directly — verified
+// on device (Campo → CoreSpotlight → on-device model). A named intent would be
+// redundant, and "Ask <app>" collides with Siri's built-in "Ask". The in-app
+// `AskView` remains as a manual on-device-RAG surface.
+
 // MARK: - App Shortcuts
 
 struct MarksShortcuts: AppShortcutsProvider {