Add OpenAI.rough_token_count class method

[jamiemccarthy]

It can be useful to estimate the token count of some text before sending it to OpenAI. This PR adds an OpenAI.rough_token_count class method which does this.

To perform the estimation, it uses the two "helpful rules of thumb" in OpenAI's docs and takes the arithmetic average.

The empty string returns 0, and any other string returns an integer greater than 0.

This is a rewrite of #306 to remove the Tiktoken functionality and change the interface as requested.

All Submissions:

• Have you followed the guidelines in our Contributing document?
• Have you checked to ensure there aren't other open Pull Requests for the same update/change?
• Have you added an explanation of what your changes do and why you'd like us to include them?

[alexrudall]

This is great, thanks a lot for rewriting. Just some documentation needed, would you mind adding a section to the README with a couple lines briefly explaining how to estimate tokens in a string and link to the OpenAI docs. Probably between Azure and Models sections I think?

And couple inline comments also, just with the method.

[alexrudall]

This is perfect thanks @jamiemccarthy!