Fix the handling of trailing junk with Content-Encoding

Fix the handling of trailing junk with Content-Encoding

When a content-encoded response has trailing junk,
SoupConverterWrapper needs to make sure that all of that junk gets
read before returning G_CONVERTER_FINISHED, or else it will end up
becoming the start of the next response on that connection.

coding-test wasn't hitting this, because the trailing junk was always
small enough that it would have been read into GConverterInputStream's
internal buffer already at the point when the GZlibDecompresser
returned FINISHED. Switching coding-test to use chunked encoding
caused it to reliably hit the bug though, because the final 0-length
chunk (including its header) will always still be unread at that
point. So update it to use that.

Also add a SoupRequest-based test to coding-test, though it turns out
to not have been relevant to this bug, which affected the SoupMessage
API too.

https://bugzilla.gnome.org/show_bug.cgi?id=676477