chore: revert model back to Qwen2.5-1.5B for CPU performance and fix iOS microphone deactivation

feat: upgrade GGUF fallback model to Qwen2.5-3B and implement dynamic tool calling agent loop

fix: optimize SpeechRecognition for iOS Safari/Chrome compatibility

style: remove sidebar chat icon and beautify chat item text styling

feat: add circular favicon using logo.svg

Implement voice chat, round send button, and image generation mode

Adjust mobile header padding to clear status bar

Deploy latest frontend and Firebase updates to Hugging Face

Add rich request logging and timers to server

Optimize Hugging Face inference with Qwen 1.5B GGUF and n_threads=2

Configure strict binary installation for llama-cpp-python

Revert Dockerfile to cached apt layer since we use pre-compiled wheels

Use pre-compiled llama-cpp-python wheel to prevent build timeout

Add g++ and make to Dockerfile for building llama-cpp-python

Switch entirely to llama-cpp-python for fast GGUF inference everywhere

Configure robust Wikipedia search fallback and Ollama-ready support

Configure Hugging Face Space metadata in README.md

Optimize server response speed and smart routing

Configure Firebase Hosting and smart API URL routing

Clean commit of Omerta AI project for deployment