Revert Whisper build download - causing OOM (exit 137). Will use lazy loading on first request.

Dockerfile

@@ -35,11 +35,13 @@ RUN python -c "from transformers import AutoTokenizer, AutoModelForSequenceClass
     AutoModelForSequenceClassification.from_pretrained('Cyberlace/swara-structure-model', cache_dir='/.cache'); \
     print('✅ Structure Model cached!')" && chmod -R 777 /.cache
 
-# 2. Download Whisper medium model (~1.5GB) - with memory optimization
-RUN python -c "import whisper; \
-    print('📥 Downloading Whisper medium model (1.5GB)...'); \
-    whisper.load_model('medium', download_root='/.cache'); \
-    print('✅ Whisper medium cached!')" && chmod -R 777 /.cache
+# 2. Whisper medium will be downloaded on FIRST REQUEST (lazy loading)
+# Build OOM with 1.5GB download - HF Space has limited build memory
+# First request will take ~2-3 min extra for one-time download
+# RUN python -c "import whisper; \
+#     print('📥 Downloading Whisper medium model (1.5GB)...'); \
+#     whisper.load_model('medium', download_root='/.cache'); \
+#     print('✅ Whisper medium cached!')" && chmod -R 777 /.cache
 
 # 3. Download Sentence Transformer for Keywords (~420MB)
 RUN python -c "from sentence_transformers import SentenceTransformer; \