Add guideline docs for creating new dialects

[h4v0kh3l1]

[dlthomas]

@etrabelsi In response to @h4v0kh3l1's comment (https://github.com/uber/queryparser/issues/32#issuecomment-387531103):

That sounds about right. There's also the option of working from the grammar for the dialect, if you have it, as Matt did for Presto.

With either approach, I'd like to stress the importance of having a corpus of valid queries - the bigger the better.