Continuously retry on retriable exceptions

This is a reasoanble thing for users to expect and IO.select docs
even have an example that do this for IO::WaitReadable.

Closes ruby-amqp#548.

lib/bunny/cruby/socket.rb

@@ -59,11 +59,8 @@ def read_fully(count, timeout = nil)
         # @eof will break Rubinius' TCPSocket implementation. MK.
         @__bunny_socket_eof_flag__ = true
       rescue *READ_RETRY_EXCEPTION_CLASSES
-        if IO.select([self], nil, nil, timeout)
-          retry
-        else
-          raise Timeout::Error, "IO timeout when reading #{count} bytes"
-        end
+        IO.select([self], nil, nil, timeout)
+        retry
       end
       value
     end # read_fully

lib/bunny/cruby/ssl_socket.rb

@@ -44,20 +44,14 @@ def read_fully(count, timeout = nil)
           @__bunny_socket_eof_flag__ = true
         rescue OpenSSL::SSL::SSLError => e
           if e.message == "read would block"
-            if IO.select([self], nil, nil, timeout)
-              retry
-            else
-              raise Timeout::Error, "IO timeout when reading #{count} bytes"
-            end
+            IO.select([self], nil, nil, timeout)
+            retry
           else
             raise e
           end
         rescue *READ_RETRY_EXCEPTION_CLASSES => e
-          if IO.select([self], nil, nil, timeout)
-            retry
-          else
-            raise Timeout::Error, "IO timeout when reading #{count} bytes"
-          end
+          IO.select([self], nil, nil, timeout)
+          retry
         end
         value
       end