Fix infinite loop when the connection closes partway through a bulk reply - #1731

fread() gives back an empty string at EOF rather than false, so when the connection closes partway through a bulk reply, predis keeps waiting for bytes it'll never recieve and pegs a core until you kill it. read_write_timeout doesn't help because every read returns straight away.

You don't need anything exotic to hit it. A pubsub subscriber that falls behind gets cut off by Redis's default client-output-buffer-limit pubsub 32mb 8mb 60, and when the cut lands in middle of a message the subscriber on 3.6.1 just spins. Killing Redis during a big GET does the same.

Both connection classes had the same loop, so both now throw the usual ConnectionException when a read comes back empty at EOF, same as a close between replies already does. The tests are most of the diff, the fix itself is a few lines in each class.