Fix language override ignoring cache; fix BGE-M3 position_ids error

Language fix:
- When --language is explicitly passed, auto-enable force-regenerate
  so cached English sections are not reused for Chinese notes (and
  vice versa). Previously, selecting Chinese produced English notes
  because existing cached section .md files were served from disk
  without checking their language.

BGE-M3 fix:
- Add trust_remote_code=True when loading BAAI/bge-m3 via
  SentenceTransformer in suggest_matches(). This fixes the
  "embeddings.position_ids status unexpected" error that occurs
  with newer sentence-transformers versions.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: nodeeeeee

note_generation.py

@@ -1593,6 +1593,10 @@ def main() -> None:
     if args.language:
         NOTE_LANGUAGE = args.language
         print(f"Note language: {NOTE_LANGUAGE}")
+        # Changing language invalidates cached sections — auto-enable force
+        if not args.force:
+            args.force = True
+            print(f"  (force-regenerate enabled: language differs from cached sections)")
 
     if args.course:
         course_dir  = COURSE_DATA_DIR / args.course

semantic_alignment.py

@@ -1486,7 +1486,8 @@ def suggest_matches(
             import torch
             device = "cuda" if torch.cuda.is_available() else "cpu"
             print(f"  [match] Loading {model_name} on {device}...")
-            embedder = SentenceTransformer(model_name, device=device)
+            embedder = SentenceTransformer(model_name, device=device,
+                                           trust_remote_code=True)
             all_texts = cap_texts + slide_texts
             embs = embedder.encode(
                 all_texts,