fix debug logger output (#458)

our dear friend mr. funclogger was bypassing calls to our multi writer since
we were embedding a *bytes.Buffer, it was using ReadFrom and WriteString which
would never call the stderr logger's Write method (or, as I learned, other
things trying to wrap that buffer's Write method...).

the tl;dr is many times DEBUG lines don't get spat out, from async tasks
especially (few people using this).

I think the final solution is probably to make funclogger a 'more robust'
interface that we understand instead of trying to minimize it to an
io.ReaderWriterCloser, much like how bytes.Buffer has all kinds of
methods implemented on it, we can implement things like ReadFrom and
WriteString most likely. not a big fan of how things are now (and it's my own
doing) with the readerwritercloser coming from multiple places but meh,
will get to it some day soon, the log stuff will be a pretty hot path.

api/agent/func_logger.go

@@ -48,15 +48,20 @@ func setupLogger(logger logrus.FieldLogger) io.ReadWriteCloser {
 	return &rwc{mw, dbuf}
 }
 
-// implements io.ReadWriteCloser, keeps the buffer for all its handy methods
+// implements io.ReadWriteCloser, fmt.Stringer and Bytes()
+// TODO WriteString and ReadFrom would be handy to implement,
+// ReadFrom is a little involved.
 type rwc struct {
 	io.WriteCloser
-	*bytes.Buffer
+
+	// buffer is not embedded since it would bypass calls to WriteCloser.Write
+	// in cases such as WriteString and ReadFrom
+	b *bytes.Buffer
 }
 
-// these are explicit to override the *bytes.Buffer's methods
+func (r *rwc) Read(b []byte) (int, error) { return r.b.Read(b) }
-func (r *rwc) Write(b []byte) (int, error) { return r.WriteCloser.Write(b) }
+func (r *rwc) String() string             { return r.b.String() }
-func (r *rwc) Close() error                { return r.WriteCloser.Close() }
+func (r *rwc) Bytes() []byte              { return r.b.Bytes() }
 
 // implements passthrough Write & closure call in Close
 type fCloser struct {