Add badResponseLoggingWriter.Flush

[leizor]

This PR adds Flush() to the middleware.badResponseLoggingWriter, making it implement http.Flusher.

I think, ideally, none of this middleware should interfere with what the ultimate HTTP handlers want to try to do. Making a best-effort attempt at Flush() seems like the friendliest behavior for middleware; if it can flush, great, otherwise nothing happens.

Additionally, I think this best-effort type behavior plays nicely with this disclaimer about http.Flusher:

Note that even for ResponseWriters that support Flush, if the client is connected through an HTTP proxy, the buffered data may not reach the client until the response completes.

[bboreham]

It looks to me like this method will advertise the possibility of Flush() on writers that don't support it. Even though there is a disclaimer, this could break something. Hyram's Law.

What about this: create a struct embedding badResponseLoggingWriter that implements Flusher, and select which one to create in newBadResponseLoggingWriter.

[leizor]

I like that much better - will do!

[leizor]

@bboreham I think this is ready for a second look