Add Weight Decay to optimizers

- [x] Sgd `weight_decay: Option<f32>`
- [ ] RMSprop `weight_decay: Option<f32>`
- [ ] Adam/AdamW: `weight_decay: Option<AdamDecay>`, where `enum AdamDecay { Vanilla(f32), AdamW(f32) }`.

(maybe use different terms for `AdamDecay`).

