Added subgradient definition and properties

[ISIPINK]

Builds on PR LeanMachineLearning#250, ISIPINK/LML branch Bregman

Added subgradient definition and properties

Design Decisions

• carrying V: subgradient is defined over V to avoid indicator functions ie $g \in \partial [y](f + I_V) \iff g \in \partial [V, y] f $
• 0 ≤ D_[f](x, y, g_y) vs classical def: the classic def needs ordered group to be equivalent, other subgradients like Clarke, Dini are not needed for Online learning and arent taken in consideration here