Bug with trimming padded array in fft_resample

[Pezz89]

Calling:
raw, events = raw.resample(128, events=events, npad=10)

Results in the following error:

Traceback (most recent call last):
  File "./pyaepsin.py", line 359, in <module>
    main(**args)
  File "./pyaepsin.py", line 178, in main
    raw, events = raw.resample(128, events=events, npad=10)
  File "<string>", line 2, in resample
  File "/Users/sam/Work/SOTON/Initial_work/TRF_Verification/mne-python/mne/utils.py", line 724, in verbose
    return function(*args, **kwargs)
  File "/Users/sam/Work/SOTON/Initial_work/TRF_Verification/mne-python/mne/io/base.py", line 1522, in resample
    window=window, n_jobs=n_jobs, pad=pad))
  File "<string>", line 2, in resample
  File "/Users/sam/Work/SOTON/Initial_work/TRF_Verification/mne-python/mne/utils.py", line 724, in verbose
    return function(*args, **kwargs)
  File "/Users/sam/Work/SOTON/Initial_work/TRF_Verification/mne-python/mne/filter.py", line 1600, in resample
    cuda_dict, pad)
ValueError: could not broadcast input array from shape (0) into shape (100224)

Due to incorrect handling of array trimming when trimming the end of the array is 0 samples long in cuda.py (line 378-383):

    # now let's trim it back to the correct size (if there was padding)
    if (to_removes > 0).any():
        keep = np.ones((new_len), dtype='bool')
        keep[:to_removes[0]] = False
        keep[-to_removes[1]:] = False
        y = np.compress(keep, y)