Escape prompt prefix/suffix along with prompt

[pnb]

I've been using some models that are a bit picky about the prompt format during chats. In particular, it seems like some of them prefer a newline after the reverse prompt (e.g., ASSISTANT:\n). The existing "main" example does allow parsing newlines and other escape characters in the prompt with -e, but not in the prompt prefix or suffix (--in-prefix and --in-suffix). This pull request adds escaping to those parts as well, allowing for things like this:

./main -m some-model.bin -f ./prompts/some-prompt.txt -i -e --reverse-prompt 'USER:' --in-prefix ' ' --in-suffix 'ASSISTANT:\n'

This inserts "ASSISTANT:" and a newline after the user presses enter, which works a bit better in some of my cases. I haven't needed escape characters in the prompt prefix, but it seems better to parse them anyway for consistency. Possibly there are use cases where it would be needed that I haven't encountered.

[wtarreau]

Ah, nice, thanks for this one, I've been annoyed several times as well.