Checkpoint error with --is_train on celebA dataset

After fixing #109 trying to train leads to a failure to find checkpoints.
It seems related to #82 but the fixes for that have been completed.

Any thoughts?

```
$ python main.py --dataset celebA --input_height=108 --is_train --is_crop True
....stuff....
[*] Reading checkpoints...
[*] Failed to find a checkpoint
[!] Load failed...
[*] 0
[*] 1
[*] 2
[*] 3
......
```

Start by reproducing the command in main.py with the celebA dataset and inspect the checkpoint-reading step after the “Reading checkpoints...” message. Compare the behavior with issues #109 and #82; done means training finds or handles the expected checkpoint instead of reporting “Load failed.”

Written by the indexing model from the issue text.