Implement locally connected layer (1-d)

Like convolutional, except that each window has its own set of filters (rather than being convolved into one set of filters).

Keras reference: https://www.tensorflow.org/api_docs/python/tf/keras/layers/LocallyConnected1D