loading (partial) prompt from files

When testing different settings or loading more complex input you usually will not want to use the -p "PROMPT" argument.

ggllm.cpp allows to load a prompt from a text -f /path/to/file and combine that with additional features such as:

• -enc : All your prompt will be encapsulated by the correct tokens/syntax for your model type.
  So for OpenAssistant V2.5 it's <|prompter|><|assistant|> for Wizard it's ### Response: etc.
  This feature relies on the model having special vocabulary (OA 2.5), using a good path or filename (oasst1) or using --alias
• -sys and -sysraw : This puts a system prompt before all other prompts. -sys will follow -enc syntax, -sysraw will not add anything to it.
  The system prompt differs from a normal prompt as it will stay on top of the context window even if your generated output exceeds -c ctx
  Example: -n 32000 -enc -sys "Write a story about a woman in the forest, searching for her lost daughter but getting harassed by wild rabbits and bugs" --ignore-eos
  This example will write 32000 tokens of a story about that woman searching for her kid, it should stay focused on that plotline even when context is exceeded.
• -e : This will change all literal "\n" into new lines in your prompts, allows to add new lines easily.
• -p "PROMPT" : This allows to add custom strings before or after the file you specified.
  Example -e -p "Summarize this code:\n" -f "sourcecode.txt" -p "\nSummarize it now please, focus on xxx." -enc
  This example will construct a 3 part prompt with new lines and encapsulte the entire combination into a finetune instruction.