WIP: SuperDataset Loader and UI Overhaul

[d8ahazard]

So, just wanted to show off what I'm currently working on, which should hopefully be the last big update like this for a while.

For starters, I've made my own version of the DreamboothDataset class that I'm affectionately calling a "SuperDataset", because when I'm done with it, it should hopefully clean a lot of stuff up and make things a lot smarter.

Based on requests to allow more control over training multiple concepts at once, I've re-worked the dataset loader so that each individual concept can have a "max steps" parameter set, which can be used to prevent over-training when training multiple concepts with varying numbers of input images.

Similarly, each concept will now have 100% control over all class and sample generation params, so you don't have to have the same number of class images per concept, and every [filewords] prompt can be constructed as-needed.

In addition, I'm adding these multiple concept options to the UI, so you don't have to mess around with JSON. Just pick the tab, fill in the fields, let the program do it's thing.

On the backend, I've updated the DreamboothConfig class so that it's a proper class now, not just a hacky dictionary. This will ensure that I don't have to do a bunch of dumb stuff every time I add a new parameter. And on the UI side, there's one method that saves input params, not a dozen, so less work for me there too. You do a thing, the config gets saved if applicable, then the called method just loads the config, versus trying to pass a billion parameters.

What I'm looking for from everybody else is any other suggestions while I'm doing this stuff that could be applicable. Anything else I'm not thinking of in regards to handling multiple concepts, etc.

I've got the WIP code here for now if anybody wants to take a look. Be warned, it's still not done, and most things are broken right now, but at least you can see the madness I'm undertaking:

https://github.com/d8ahazard/sd_dreambooth_extension/tree/SuperDataset