Update: Live streaming, working config.json layer, latency tuning

[nidhiii128]

Update: Live streaming + working config.json layer + latency tuning

Streaming (text + vision):

• OllamaWorker and OllamaVisionWorker now emit each token via a new
  chunk_signal.
• Chatbot.py consumes it (_on_stream_chunk) and rewrites the bot bubble
  in place using an anchored cursor, so replies render token-by-token instead
  of appearing all at once. Generation is interruptible via the Stop button.

config.json (customizable system rules):

• chatbot_thread.py now loads src/chatbot/config.json at startup and uses
  it for the system prompt, context window, sampling params, keep-alive, and
  history depth. Editing the file changes Copilot behavior without code edits;
  falls back to built-in defaults if the file is missing or invalid.

Latency tuning:

• num_ctx 2048 → 1024, history window 10 → 6 turns, lower token-budget tiers
  (128/256/512), and keep_alive set to never-unload so repeat questions skip
  the model reload cost.