Update model_probe to work with diffuser-format SD TI embeddings. (#5301)

## What type of PR is this? (check all applicable)

- [ ] Refactor
- [ ] Feature
- [x] Bug Fix
- [ ] Optimization
- [ ] Documentation Update
- [ ] Community Node Submission

      
## Have you updated all relevant documentation?
- [x] Yes (N/A)
- [ ] No


## Description

This change enables the model probe to work with TI embeddings that have
the follow state_dict structure:

```python
{
    "<any_key>": torch.Tensor(...), # where the tensor has shape (N, embedding_dim)
}
```

## QA Instructions, Screenshots, Recordings

I can't imagine an embedding format that would previously have passed
the model probe, and would now fail after this change. That being said,
I'll exercise a bunch of existing TIs before merging.

- [x] Exercise existing TI formats


## Added/updated tests?

- [ ] Yes
- [x] No : _We could really benefit from tests for all of the supported
TI formats... but I'm not taking on that project right now._

invokeai/backend/model_management/model_probe.py

@@ -389,7 +389,7 @@ class TextualInversionCheckpointProbe(CheckpointProbeBase):
         elif "clip_g" in checkpoint:
             token_dim = checkpoint["clip_g"].shape[-1]
         else:
-            token_dim = list(checkpoint.values())[0].shape[0]
+            token_dim = list(checkpoint.values())[0].shape[-1]
         if token_dim == 768:
             return BaseModelType.StableDiffusion1
         elif token_dim == 1024: