This caused a lot of re-rendering whenever the selection changed, which caused a huge performance hit. It also made changing the current image lag a bit.
Instead of providing an array of image names as a multi-select dnd payload, there is now no multi-select dnd payload at all - instead, the payload types are used by the `imageDropped` listener to pull the selection out of redux.
Now, the only big re-renders are when the selectionCount changes. In the future I'll figure out a good way to do image names as payload without incurring re-renders.
Every `GalleryImage` was rerendering any time the app rerendered bc the selector function itself was not memoized. This resulted in the memoization cache inside the selector constantly being reset.
Same for `BatchImage`.
Also updated memoization for a few other selectors.
Eg `useGetMainModelsQuery()`, `useGetLoRAModelsQuery()` instead of `useListModelsQuery({base_type})`.
Add specific adapters for each model type. Just more organised and easier to consume models now.
Also updated LoRA UI to use the model name.
This PR is for adjusting the unit tests in the `tests` directory so that
they no longer throw errors.
I've removed two tests that were obsoleted by the shift to latent nodes,
but `test_graph_execution_state.py` and `test_invoker.py` are throwing
this validation error:
```
TypeError: InvocationServices.__init__() missing 2 required positional arguments: 'boards' and 'board_images'
```
The `invokeai-configure` script migrates the old invokeai.init file to
the new invokeai.yaml format. However, the parser for the invokeai.init
file was missing the names of the k* samplers and was giving a parser
error on any invokeai.init file that referred to one of these samplers.
This PR fixes the problem.
Ironically, there is no longer the concept of the preferred scheduler in
3.0, and so these sampler names are simply ignored and not written into
`invokeai.yaml`
