fix conv1d channels first

[roywei]

• Fix #122 and #121, tested with unit tests and end to end example (imdb_cnn.py)
• Fix Conv1D implementation to change kernel shape according to image data format
• Change default Conv1D data format to None (use settings in keras.json), which is consistent with other Conv Layers (raised issue with keras-team/keras, will follow up with PR if needed)
• Enabled test_conv_1d to test channels_first data format. While avoid testing causal with channels_first as Keras does not support.

  if padding == 'causal':
          if data_format != 'channels_last':
              raise ValueError('When using causal padding in `Conv1D`, '
                               '`data_format` must be "channels_last" '
                               '(temporal data).')

[roywei]

Merging, updated changes made on Keras code #92