Support ShrinkingSplitter

[eromoe]

I didn't add correct comment, since don't know if you could appove this.
I came up this for how different lengths of train data affect on same testing set, so that I can choose a good realtime train length.

import numpy as np

a = pd.DataFrame(np.arange(0, 30).reshape(-1, 3), columns=[1,2, 3])
a.vbt.split(ShrinkingSplitter(), n=5, min_len=6, set_lens=(3,), plot=True, left_to_right=False, trace_names=['in-sample', 'out-sample']).show()

[catbrower]

why is this PR