Add Dropout to deeper/wider unet

[alxmrs]

Drop out could improve the ability of the model to converge as we make it larger.

[alxmrs]

• Need to re-run with new dataset (unfiltered) --> more experiments needed.

[alxmrs]

We are experimenting with is right now! xref: #661 #641 #642 #733

I am not closing this issue because there are a lot of types of dropout we could use. If you're a newcomer to this project and want to take this on, I recommend starting with adding layer-wise drop-out to our ConvNeXt blocks or "traditional" stochastic depth dropout in our U-Net.

[alxmrs]

@amogh-gulati Sorry, I just noticed that you're assigned to this issue. I'm going to create new sub-issues here to track this for newcomers.