RequestSequence is difficult to use on Python 3

[twm]

RequestSequence is very picky about types of arguments, but lacks any checking or coercion. For example, in the request tuple:

• b'get' vs. 'get'—the former is required
• b'get' vs. 'GET'—the former is required
• The URL is str, but everything else must be bytes

If you pass the wrong type, the request sequence will not ever match. I'd say we should blacklist str where bytes is required and vice-versa. (Blacklist rather than whitelist because the ability to pass fuzzy matchers like HasHeaders is useful.)

In the response tuple:

• Headers must be a dict of strings, as noted in StringStubbingResource docstring is wrong #123, which is inconsistent with HasHeaders. HasHeaders requires lists of strings.

And generally speaking, writing out these tuples is easy to botch. Perhaps we should add a builder API to make it easier?