Factor out some functions into llm-utils

[nicovank]

This is just a proposal to factor out some functions into a common repository, for CWhy only for now.

I've used https://github.com/plasma-umass/llm-utils.

Transferred functions:

• count_tokens
• calculate_cost: This is not used by CWhy yet.
• word_wrap_except_code_blocks
• read_lines: This function is slightly different in ChatDBG, but it shouldn't be hard to make it compatible.

Future candidates:

• complete: Or at least some version if call GPT, print error if missing OPEN_API_KEY or other error.
• Support for other LLMs (Llama?)

Thoughts? Do we like it? Is it worth it?