Add helper function applying correct padding for even kernel sizes

[DanEnergetics]

It was mentioned in #50 and I tested it as well (see test case), the get_same_padding doesn't seem to work for even kernel sizes. It applies the same padding (k - 1)//2 on both sides in a given dimension, which is correct whenever (k-1) is even, otherwise a frame is missing. However, sometimes it does seem to work if stride > 1.
apply_same_padding is to be used as in the test case by first padding the tensor and then calling a convolution without any padding on it.