Move note language computation off the main thread

NLLanguageRecognizer.processString() is an expensive NLP operation that
was moved to @MainActor in 5058fb33, causing UI jank when scrolling.

This moves language detection back to the async preload path where it
runs off the main thread. get_preload_plan is synchronous again and
defers language computation to preload_event. Translations still happen
on the first preload pass since preload_event now checks .havent_tried
directly rather than relying on the load_translations flag from the plan.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

nostrdb/NdbNote.swift

@@ -568,8 +568,8 @@ extension NdbNote {
         return thread_reply() != nil
     }
 
-    @MainActor
-    func note_language(ndb: Ndb, _ keypair: Keypair) -> String? {
+    func note_language(_ keypair: Keypair) -> String? {
+        assert(!Thread.isMainThread, "This function must not be run on the main thread.")
 
         // Rely on Apple's NLLanguageRecognizer to tell us which language it thinks the note is in
         // and filter on only the text portions of the content as URLs and hashtags confuse the language recognizer.