[Feature Request] Add multichannel support to RandomShadow

[ternaus]

RIght now we have:

mask = np.zeros_like(img, dtype=np.uint8)
cv2.fillPoly(mask, vertices_list, (max_value, max_value, max_value))

in add_shadow function.

cv2.fillPolly, does not work with the number of channels > 4

=>

1. if number of channels is larger than 4, create mask for 1 channel and replicate it to the desired number of channels.
2. check that everything works for grayscale image
3. tests
4. update docstrings in add_shadow and RandomShadow