memory optimization ready :D

[furllmm]

Optimizations Implemented:

1. Lazy Loading for Image Embeddings:

   • Startup Optimization: Removed the proactive loading of the image embedding model in webui.py . The model will now only load when you actually navigate to the Search tab and use "Image Semantic Search" or "Similar Image Search".
   • Smart Caching: Modified windrecorder/ui/components.py to check if the model is already loaded before attempting to load it again, preventing redundant memory usage.

2. Memory Cleanup Controls:

   • Image Embedding Model: Added an unload_model function in windrecorder/img_embed_manager.py to explicitly release memory occupied by the text/image encoders and processors.
   • OCR Engines: Added an unload_ocr_engine function in windrecorder/ocr_manager.py to release memory used by PaddleOCR, ChineseOCR Lite, or WeChatOCR when they are no longer needed.

3. New "Memory Management" Section:

   • Added a Memory Management section in the Setting tab.
   • "Unload Embedding Models" Button: Frees up RAM by unloading the image embedding models.
   • "Unload OCR Engines" Button: Frees up RAM by unloading any active OCR engines.
     These changes ensure that the application consumes significantly less RAM on startup and gives you control to free up resources after performing heavy tasks like indexing or searching.