fix(encoding): Fix double-encoding of HTTP response (#220)

[spinnakerbot]

ecba3ea added a call to encode
the response body before sending it to the server if the body is
of type str.

This is the correct behavior for python 3, where str represents a
an unencoded string, but is incorrect for python 2. In python 2,
str represents an already-encoded string and unicode represents
an unencoded string.

This is breaking for prometheus monitoring, which is sending already-
encoded bytes to the respond() function. In python 2, those have type
str, so we're trying to re-encode them. This works fine as long as there
are only ASCII characters in the string, but breaks otherwise.

In order to support both python 2 and 3, try to use 'unicode' (as
exists in 2) and if it's absent assign str to unicode.

Automated cherry pick of 59cbbec into 1.16