bugfix: Float64 error for mps devices on set_timesteps (#4040)

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

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


## Have you discussed this change with the InvokeAI team?
- [ ] Yes
- [x] No, because: minor fix, let me know your thoughts

      
## Have you updated all relevant documentation?
- [x] Yes
- [ ] No

## Description


## Related Tickets & Documents

<!--
For pull requests that relate or close an issue, please include them
below. 

For example having the text: "closes #1234" would connect the current
pull
request to issue 1234.  And when we merge the pull request, Github will
automatically close the issue.
-->

- Related Issue # https://github.com/invoke-ai/InvokeAI/issues/4017
- Closes #

## QA Instructions, Screenshots, Recordings

<!-- 
Please provide steps on how to test changes, any hardware or 
software specifications as well as any other pertinent information. 
-->

## Added/updated tests?

- [ ] Yes
- [x] No : Requires mps device

## [optional] Are there any post deployment tasks we need to perform?

Please test on an MPS (M1/M2) device. 

Relevant code causing the error in #4017 


01b6ec21fa/src/diffusers/schedulers/scheduling_euler_discrete.py (L263C3-L268C75)

```
        self.sigmas = torch.from_numpy(sigmas).to(device=device)
        if str(device).startswith("mps"):
            # mps does not support float64
            self.timesteps = torch.from_numpy(timesteps).to(device, dtype=torch.float32)
        else:
            self.timesteps = torch.from_numpy(timesteps).to(device=device)
```

invokeai/app/invocations/sdxl.py

@@ -292,15 +292,16 @@ class SDXLTextToLatentsInvocation(BaseInvocation):
         )
 
         num_inference_steps = self.steps
-        scheduler.set_timesteps(num_inference_steps)
-        timesteps = scheduler.timesteps
-
-        latents = latents * scheduler.init_noise_sigma
 
         unet_info = context.services.model_manager.get_model(**self.unet.unet.dict(), context=context)
         do_classifier_free_guidance = True
         cross_attention_kwargs = None
         with unet_info as unet:
+            scheduler.set_timesteps(num_inference_steps, device=unet.device)
+            timesteps = scheduler.timesteps
+
+            latents = latents.to(device=unet.device, dtype=unet.dtype) * scheduler.init_noise_sigma
+
             extra_step_kwargs = dict()
             if "eta" in set(inspect.signature(scheduler.step).parameters.keys()):
                 extra_step_kwargs.update(
@@ -537,27 +538,28 @@ class SDXLLatentsToLatentsInvocation(BaseInvocation):
             scheduler_name=self.scheduler,
         )
 
-        # apply denoising_start
-        num_inference_steps = self.steps
-        scheduler.set_timesteps(num_inference_steps)
-
-        t_start = int(round(self.denoising_start * num_inference_steps))
-        timesteps = scheduler.timesteps[t_start * scheduler.order :]
-        num_inference_steps = num_inference_steps - t_start
-
-        # apply noise(if provided)
-        if self.noise is not None and timesteps.shape[0] > 0:
-            noise = context.services.latents.get(self.noise.latents_name)
-            latents = scheduler.add_noise(latents, noise, timesteps[:1])
-            del noise
-
         unet_info = context.services.model_manager.get_model(
             **self.unet.unet.dict(),
             context=context,
         )
+
         do_classifier_free_guidance = True
         cross_attention_kwargs = None
         with unet_info as unet:
+            # apply denoising_start
+            num_inference_steps = self.steps
+            scheduler.set_timesteps(num_inference_steps, device=unet.device)
+
+            t_start = int(round(self.denoising_start * num_inference_steps))
+            timesteps = scheduler.timesteps[t_start * scheduler.order :]
+            num_inference_steps = num_inference_steps - t_start
+
+            # apply noise(if provided)
+            if self.noise is not None and timesteps.shape[0] > 0:
+                noise = context.services.latents.get(self.noise.latents_name)
+                latents = scheduler.add_noise(latents, noise, timesteps[:1])
+                del noise
+
             # apply scheduler extra args
             extra_step_kwargs = dict()
             if "eta" in set(inspect.signature(scheduler.step).parameters.keys()):