LlaMa Factory Playbook Improvements - making it stable for launch across all OS's and GPU devices

[ldokovic-personal]

Key Changes:

• Adding Windows OS as a filter category on UI (button)
• Inclusion of all GPU devices as a filter category on UI (buttons)
• Separating venvs when the AI Halo device
• Repair of Windows launch by adjusting 2 additional parameters and clear steps that the user follows with an explanation of why,
  1. bf16 -> in supported fp16
  2. dataloader_num_workers -> single process on Windows to avoid the impossibility of serializing pickle local object (known issue on LlamaFactory repo)
• Fixed Linux Tests and added new ones for Windows
• Improved description of LlamaFactory WebUI - for easy navigation of our users

Overall, fixing everything that is addressed in the following two reports:

1. Report on Llama factory finetuning - Linux
2. Report on Llama Factory finetuning - Windows