Support MimeParse-ing on things other than Strings, and allow mixing and matching MimeRender destination types

[iostat]

Hey!

This PR adds a couple of nifty features. One is that you can now MimeParse stuff other than Strings -- handy for when your request body might not be valid UTF-8 (e.g., is binary). Additionally, you can now mix and match different kinds of MimeRender instances -- previously, if some routes MimeRendered down to a String and others to a ByteString, for example -- you'd get a type mismatch (MimeRender a b String vs MimeRender c d ByteString) since not all of the final rendered types were the same, and the compiler picked whichever came first.

To enable the latter, I had to generalize how responses are written to the output stream (hence the ResponseWriter type), and needed to add an NPM package to facilitate testing (stream-buffers, which allow you to treat a Buffer as a node stream).

You may find there is no obvious test for the mix-and-match MimeRender functionality -- the test is the test suite itself compiling! Note that MagicContentType now MimeRenders to a ByteString, whereas there are remaining HTML and JSON endpoints which MimeRender to String:

@@ -44,8 +46,8 @@ magicMimeTypeRenderString = "It's magic, you know!"
 instance hasMediaTypeMagic :: HasMediaType MagicContentType where
   getMediaType _ = magicMimeType
 
-instance magicallyRenderMagicContentType :: MimeRender a MagicContentType String where
-  mimeRender _ _ = magicMimeTypeRenderString
+instance magicallyRenderMagicContentType :: MimeRender a MagicContentType ByteString where
+  mimeRender _ _ = ByteString.toUTF8 magicMimeTypeRenderString